Broad strokes, 2011-2026

123456789101112 A

Hover a well for the paper, click it for the card. The cards show each figure as published, and clicking a figure opens it at full size.

Each figure has a story about why I picked it. I have written only the first, A1; the others I am saving as a treat for someday.

2013

2013, Journal of Biomolecular Screening. Well A1.

Comparison of Methods for Image-Based Profiling of Cellular Morphological Responses to Small-Molecule Treatment

Vebjorn Ljosa, Peter D. Caie, Rob ter Horst, Katherine L. Sokolnicki, Emma L. Jenkins, Sandeep Daya, Mark E. Roberts, Thouis R. Jones, Shantanu Singh, Auguste Genovesio, Paul A. Clemons, Neil O. Carragher, Anne E. Carpenter

Read the paper

We compared five ways to combine single-cell measurements from microscope images into one profile per sample by how well each predicted a drug's mechanism of action, the way it acts on cells. In treated breast cancer cells, simply averaging each measurement did as well as methods modeling single cells. Applying factor analysis before averaging worked best, predicting the mechanism correctly for 94% of treatments with known mechanisms. We released the data and open-source code.

Caption

Confusion matrix for the factor-analysis method, showing the number of compound concentrations that were classified correctly (on the diagonals) and incorrectly (off the diagonals), the classification accuracies for each mechanism of action (right columns), and overall classification accuracy (number of correctly classified compound concentrations divided by the total number of compound concentrations). Average outcomes over 20 models are shown; dimly colored squares without numbers indicate classification outcomes that occurred fewer than 0.5 times on average.

Reflections

I picked the confusion matrix because it was the most evocative figure for me, in this paper and in many that followed. I remember Vebjorn had very carefully crafted its graphics using Cairo. It was my first paper at the Broad: I'd come in excited about scaling up cell imaging, had convinced Anne and Carolina to bring me on, and was thrilled to find datasets already collected. I'd been pushing these ideas during my PhD, inspired by Steve Altschuler and Lani Wu, but was constrained to hundreds of cells imaged by hand. Now we had tens of thousands, ready to go.

The confusion matrix overstates how well we could separate compounds by mechanism; our goal was simply to show that the obvious things could be done. Hard to believe now, but back then it wasn't at all clear we could distinguish compounds by phenotype. The dataset went on to become the famous BBBC021, heavily used for the next decade in image-based profiling.

Citation

Vebjorn Ljosa, Peter D. Caie, Rob ter Horst, Katherine L. Sokolnicki, Emma L. Jenkins, Sandeep Daya, Mark E. Roberts, Thouis R. Jones, Shantanu Singh, Auguste Genovesio, Paul A. Clemons, Neil O. Carragher, Anne E. Carpenter. Comparison of Methods for Image-Based Profiling of Cellular Morphological Responses to Small-Molecule Treatment. Journal of Biomolecular Screening 18, 1321-1329 (2013). https://doi.org/10.1177/1087057113503553

Back to the plate

2014

2014, Journal of Biomolecular Screening. Well A2.

Increasing the Content of High-Content Screening: An Overview

Shantanu Singh, Anne E. Carpenter, Auguste Genovesio

Read the paper

Caption

Feature set sizes used in papers throughout the three searches. Numbers at the top of each bar indicate the actual number of papers. Between 60% and 80% of the papers used only one or two measured features of the cells.

Citation

Shantanu Singh, Anne E. Carpenter, Auguste Genovesio. Increasing the Content of High-Content Screening: An Overview. Journal of Biomolecular Screening 19, 640-650 (2014). https://doi.org/10.1177/1087057114528537

Back to the plate

2014, Journal of Microscopy. Well A3.

Pipeline for illumination correction of images for high‐throughput microscopy

S. Singh, M.-A. Bray, T.R. Jones, A.E. Carpenter

Read the paper

Caption

Estimating illumination correction functions. Sample ICFs for DAPI (top-left), actin (top-right) and tubulin (bottom-left). The image dimensions are 1280 × 1024 pixels. To generate an ICF, the mean image across all images in a plate is smoothed using a median filter with window size = 500 pixels. The value of this parameter was chosen manually as follows. Start with filter dimensions approximately 25% that of the image, then increase the dimensions in increments of about 10% of the image size until the ICFs appear smooth overall; the presence of bright ‘blotches’ in the ICF indicates that the local cellular intensities still dominate the global illumination pattern or that artefacts are present. The density of cells across the field-of-view is uniform (bottom-right). This check is important to confirm that there are no systematic uneven cell distribution patterns that could lead to erroneous ICFs.

Citation

S. Singh, M.-A. Bray, T. R. Jones, A. E. Carpenter. Pipeline for illumination correction of images for high-throughput microscopy. Journal of Microscopy 256, 231-236 (2014). https://doi.org/10.1111/jmi.12178

Back to the plate

2015

2015, PLoS ONE. Well A4.

Morphological Profiles of RNAi-Induced Gene Knockdown Are Highly Reproducible but Dominated by Seed Effects

Shantanu Singh, Xiaoyun Wu, Vebjorn Ljosa, Mark-Anthony Bray, Federica Piccioni, David E. Root, John G. Doench, Jesse S. Boehm, Anne E. Carpenter

Read the paper

Caption

Visual example of seed effects dominating morphological profiles. Only nuclear shape features were used in this example in order to yield visually interpretable phenotypes. Two of the shRNA sequences targeting CDCA7L and BECN1 share a seed sequence (sequences 299864 and 17937 have a common seed GAATGA at nucleotides 12–17). The Spearman correlation between the morphological profiles of these shRNAs is high (seed correlation = 0.44, red bar). For each of these two shRNAs, a same-gene shRNA with a dissimilar phenotype is also shown (same-gene shRNA correlation = -0.31 and -0.11, green bars). All four shRNAs have high replicate correlation and are dissimilar from untreated cells. We specifically chose an example where the seed correlation is high and the same-gene shRNA correlations are low; however this phenomenon is seen globally (Fig 3). Images have been zoomed in, showing only 2% of the imaged region.

Comparison of the confusion matrices of the profiles generated using classical methods and from pretrained neural networks. The classical results were taken from Singh et al. [13]. The deep learning results were generated using the Inception-v3 architecture. Each stain was transformed as a separate greyscale image.

Examples of real BBBC021 cell images juxtaposed with synthetic cell images generated with LSGAN, WGAN and DCGAN.

Citation

Peter Goldsborough, Nick Pawlowski, Juan C. Caicedo, Shantanu Singh, Anne E. Carpenter. CytoGAN: Generative Modeling of Cell Images. bioRxiv (2017). https://doi.org/10.1101/227645

Back to the plate

2018

2018, IEEE International Symposium on Biomedical Imaging (ISBI 2018). Well B7.

Combining morphological and migration profiles of in vitro time-lapse data

Tim Becker, Juan C. Caicedo, Shantanu Singh, Markus Weckmann, Anne E. Carpenter

Read the paper

Caption

The migration profiles of all experimental groups is shown as a 360 degree histogram of the trajectories distribution. Additionally, the speed distribution in each sector is color coded similar to a windrose plot. A directed migration is represented by a dominant migration towards the chemoattractant Interleukin-8 on the left. The strongest effect of the Prednisolone treatment can be observed in Fibronectin (FN) as opposed to a collagen-rich matrix (HEM).

Citation

Tim Becker, Juan C. Caicedo, Shantanu Singh, Markus Weckmann, Anne E. Carpenter. Combining morphological and migration profiles of in vitro time-lapse data. IEEE International Symposium on Biomedical Imaging (ISBI 2018), 965-968 (2018). https://doi.org/10.1109/ISBI.2018.8363731

Back to the plate

2018, Cell Chemical Biology. Well B8.

Read the paper

Caption

Segmentation steps for the quantification of transcripts per cell within a 3D blastocyst. Images were captured of a mouse embryo blastocyst cell membrane stained with WGA and FISH for GAPDH transcripts. (A) Original 3D image of blastocyst cell membrane prior to analysis. (B) CellProfiler 3.0 image processing modules used for membrane image processing. Figure labels: RH (“RemoveHoles”), Close (“Closing”), Erode (“Erosion”), Mask (“MaskImage”), Math (“ImageMath”), EorS Features (“EnhanceOrSuppressFeatures”). (C) Nuclei after segmentation by CellProfiler, as viewed in Fiji. (D) Segmentation of cells after setting nuclei as seeds by CellProfiler, as viewed in Fiji. (E) Segmentation of GAPDH transcript foci using CellProfiler, as viewed in Fiji. (F) Examples of analysis that can be done by CellProfiler: (top) cell volume relative nucleus volume, (middle) GAPDH transcript quantity in each cell using CellProfiler’s “RelateObjects” module, (bottom) number of GAPDH transcripts in Z-plane (bin size = 2.5 μm). The underlying measurements may be downloaded as S1 File. Images were provided by Javier Frias Aldeguer and Nicolas Rivron from Hubrecht Institute, Netherlands, and are available from the Broad Bioimage Benchmark Collection (https://data.broadinstitute.org/bbbc/BBBC032/). 3D, three-dimensional; FISH, fluorescent in situ hybridization; GAPDH, glyceraldehyde 3-phosphate dehydrogenase; WGA, wheat germ agglutinin.

Accuracy and usability of segmentation strategies in the second-stage holdout sets. a, The histogram counts participant teams (n = 739) according to the official competition score of their best submission. The top five competitors are labeled in the distribution, as is the reference segmentation obtained by an expert analyst using CellProfiler. b, Accuracy of the top three solutions measured as the F1 score at multiple IoU thresholds. The scale of the x axis of the histogram in panel a (competition score) is correlated with the area under the curve of the F1 score versus IoU thresholds. The top three models had a similar performance with slight differences at the tails of the curves. c, Breakdown of accuracy in the second-stage evaluation set for the top performing model and three reference solutions. The distribution of F1-scores at a single IoU threshold (IoU = 0.7) shows points (n = 106) that each represented the segmentation accuracy of one image in the set of 106 annotated images of the second-stage evaluation (Methods). The color of single-image points corresponds to the group of images defined for reference evaluations (Methods and Fig. 2). The average of the distribution is marked with a larger point labeled with the corresponding average accuracy value. d, Estimated time required to configure the segmentation tools evaluated in c (Supplementary Note 4).

Citation

Juan C. Caicedo, Allen Goodman, Kyle W. Karhohs, Beth A. Cimini, Jeanelle Ackerman, Marzieh Haghighi, CherKeng Heng, Tim Becker, Minh Doan, Claire McQuin, Mohammad Rohban, Shantanu Singh, Anne E. Carpenter. Nucleus segmentation across imaging experiments: the 2018 Data Science Bowl. Nature Methods 16, 1247-1253 (2019). https://doi.org/10.1038/s41592-019-0612-7

Back to the plate

2020

2020, Cytometry Part A. Well C4.

Label‐Free Leukemia Monitoring by Computer Vision

Minh Doan, Marian Case, Dino Masic, Holger Hennig, Claire McQuin, Juan Caicedo, Shantanu Singh, Allen Goodman, Olaf Wolkenhauer, Huw D. Summers, David Jamieson, Frederik W van Delft, Andrew Filby, Anne E. Carpenter, Paul Rees, Julie Irving

Read the paper

Caption

Label‐free identification of ALL cells by ResNet50 corresponds well with biomarker‐based analysis. (A) Prediction results on each of 11 test data entries from five patients, who have data at the time of presentation and during treatment (average accuracy from single‐cell classification) (Test 1). The first column reports average accuracy across samples using all channels; subsequent columns correspond to incremental dropping of the next channel. BF, bright‐field; DF, dark‐field. Boxplots show the median line, first and third quartiles. Whiskers are drawn to double interquartile range (+2IQR). Diamonds represent data events that are outside the low and high whisker ends. (B) Three‐class single‐cell classification using bright‐field and dark‐field channels (Test 1). Confusion matrix for three categories, n = 218,747 ground‐truth cells pooled from 11 test sets. (C) Clustering of 9,025 cells randomly sampled from patient LK157 at the time of presentation based on deep learning features. 3D t‐Distributed Stochastic Neighbor Embedding (t‐SNE)(20) presentations based on 2,048 label‐free feature vectors for each cell from the second‐to‐last layer of the trained neural network, ResNet50, are shown. The t‐SNE calculation was stabilized after 400 iterations at perplexity of 30. Colors coded according to true class labels: Leukemic cells (blue), normal B lymphocytes (red), other cell types (yellow). The magnified images display true bright‐field channel, randomly zoomed at each cluster. Example data with overlaid images http://projector.tensorflow.org/?config=raw.githubusercontent.com/minh‐doan/Deeplearning_LabelFree_Leukemia/master/Publish/DL_supervised/Data/Step4/Output/projector_config.pbtxt can be visualized with a web‐based projector. (D) Comparison between human‐curated manual gates and label‐free deep learning in predicting residual disease fraction (Test 2). The prediction on each sample was performed on the DAPI‐positive population of in‐focus, white blood cells, which is a mixture of leukemic blasts, normal B lymphocytes, other mature and immature hematopoietic cells such as granulocytes and monocytes. Red dash‐line is 25%—the threshold of treatment effectiveness of chemotherapy. (E) Summary of residual disease estimated by deep learning using information in bright‐field channel of labeled/unlabeled samples (each with laser on/off settings) (Test 3). Residual disease fraction was calculated as the percentage of predicted leukemic cells over the whole population of unannotated in‐focus single cells. For reference, human‐curated manual residual disease fraction (last blue column) was estimated based on the stained version with laser on. (F) Result of label‐free deep learning residual disease readout (yellow columns) on unannotated in‐focus single cells in held‐out data (Test 4). Blue columns are residual disease readout reported by standard clinical flow cytometric protocol. LK881 and LK919 (asterisks) are relapsed ALL patients.

Citation

Minh Doan, Marian Case, Dino Masic, Holger Hennig, Claire McQuin, Juan Caicedo, Shantanu Singh, Allen Goodman, Olaf Wolkenhauer, Huw D. Summers, David Jamieson, Frederik W. van Delft, Andrew Filby, Anne E. Carpenter, Paul Rees, Julie Irving. Label-Free Leukemia Monitoring by Computer Vision. Cytometry Part A 97, 407-414 (2020). https://doi.org/10.1002/cyto.a.23987

Back to the plate

2020, Proceedings of the National Academy of Sciences. Well C5.

Objective assessment of stored blood quality by deep learning

Read the paper

Caption

Data-driven ordering of RBC morphologies by weakly supervised learning allows robust blood quality assessment. (A–C) We discovered a relatively linear progression for major morphological classes of RBCs using features extracted from an intermediate layer (Res4a_ReLU) of the trained weakly supervised model projected into low-dimensional space using a UMAP algorithm. This continuum is observed in 3D (A), 2D (B), or 1D (C) embedding space and interactive 3D-PCA, t-SNE, and UMAP projections of 7,000 representative cells can be explored in a public browser-based tool (ref. 36; select colors and labels for better visualization). Color coding in A, B, and C is consistent, showing that the extracted weakly supervised features place cells along their correct biological progression, from discocytes (smooth discs, crenated discs) to echinocytes (crenated discoids, crenated spheroids, crenated spheres) to spherocytes (smooth sphere). The boxes in A overlap due to continuous transitions between morphology categories, which could not be further resolved by the chosen ResNet50 architecture. The red dotted line in C indicates the threshold in the 1D UMAP, above which RBCs were categorized as unhealthy; this includes most spheroechinocytes (crenated spheroid, crenated spheres, and smooth spheres). The increasing fraction of unhealthy cells (x) over the total number of cells is termed SMI. (D) Distribution of unannotated cells according to the 1D UMAP of weakly supervised features. For each blood unit, deep-learning features were extracted from label-free images of 20,000 cells by a trained weakly supervised neural network. The extracted features were then projected in 1D UMAP space (x axis of each histogram). The shift of distributions from the left to the right as time progresses is clearly visible (more healthy biconcave RBCs are toward the negative end of the x axis; spheroechinocytes are toward the positive end). (E–H) Results from different approaches for evaluating the quality of blood units, with the y axes unified to the same scale. (E) Blood quality according to the proposed SMI. (F) Blood quality according to our automated MI morphology analysis using a fully supervised classifier. (G) Blood quality as assessed by a physiological assay for hemolysis. (H) Blood quality as assessed by expert manual MI morphology analysis (∼4,000 cells per blood unit per time point). (I–L) Pairwise comparisons between proposed machine learning approaches and classic methods for evaluating the quality of red cell units. (I) There is a stronger correlation between the proposed weakly supervised-based quality assessment and hemolytic readouts (coefficient of determination, R² = 0.93) than that of (J) human manual annotations of morphology (R² = 0.67). (K and L) In contrast, the proposed supervised learning-based method shows the opposite trend. The x and y axes of plots I–L were unified to the same scale.

Citation

Minh Doan, Joseph A. Sebastian, Juan C. Caicedo, Stefanie Siegert, Aline Roch, Tracey R. Turner, Olga Mykhailova, Ruben N. Pinto, Claire McQuin, Allen Goodman, Michael J. Parsons, Olaf Wolkenhauer, Holger Hennig, Shantanu Singh, Anne Wilson, Jason P. Acker, Paul Rees, Michael C. Kolios, Anne E. Carpenter. Objective assessment of stored blood quality by deep learning. Proceedings of the National Academy of Sciences 117, 21381-21390 (2020). https://doi.org/10.1073/pnas.2001227117

Back to the plate

2020, Cell Reports. Well C6.

TFEB Transcriptional Responses Reveal Negative Feedback by BHLHE40 and BHLHE41

Kimberly L. Carey, Geraldine L.C. Paulus, Lingfei Wang, Dale R. Balce, Jessica W. Luo, Phil Bergman, Ianina C. Ferder, Lingjia Kong, Nicole Renaud, Shantanu Singh, Maria Kost-Alimova, Beat Nyfeler, Kara G. Lassen, Herbert W. Virgin, Ramnik J. Xavier

Read the paper

Caption

BHLHE40 and BHLHE41 Repress Expression of Select TFEB Target Genes through a Negative Feedback Loop. (A) Volcano plots depicting differential gene expression in BHLHE40/41-WT versus BHLHE40/41-KO reconstituted cells at steady state. Highlighted in red (left panel) are genes significantly upregulated in TFEB-WT versus TFEB-KO cells following Torin treatment (logFC > ln4 and q < 0.01). Highlighted in cyan (right panel) are select TFEB target genes. (B–E) tSNE plots representing all RNA sequencing datasets in our study. Comparison of RNA sequencing datasets from (B) TFEB-WT and TFEB-KO cells or (C) BHLHE40/41-WT and BHLHE40/41-dKO cells treated with Torin, in which each gene (dot) is colored by the logFC. (D) Known TFEB target genes or (E) strongest putative TFEB and BHLHE40/41 competing target genes are highlighted in red. (F) Gene expression of select TFEB and BHLHE40/41 target genes in TFEB-KO, TFEB-WT, BHLHE40/41-dKO, and BHLHE40/41-WT cells as quantified by qRT-PCR. By two-tailed t test analyses, all genes were significantly upregulated in cells reconstituted with TFEB and downregulated in those reconstituted with BHLHE40/41. Data are representative from three independent experiments (mean ± SEM [standard error of the mean]). **p < 0.006, ****p < 0.0001. (G) Venn diagram represents the number of the 128 strongest putative competing target genes (logFC > ln2 and < −ln2 in TFEB- and BHLHE40/41-WT versus TFEB- and BHLHE40/41-KO cells, respectively, and q < 0.05 in both) bound by TFEB, BHLHE40, and/or BHLHE41 based on the published ChIP-seq datasets GEO: GSM2354032, GSE106000, GSM2797493, and GSM2461743. See also Figure S6 and Tables S7 and S8.

Citation

Kimberly L. Carey, Geraldine L. C. Paulus, Lingfei Wang, Dale R. Balce, Jessica W. Luo, Phil Bergman, Ianina C. Ferder, Lingjia Kong, Nicole Renaud, Shantanu Singh, Maria Kost-Alimova, Beat Nyfeler, Kara G. Lassen, Herbert W. Virgin, Ramnik J. Xavier. TFEB Transcriptional Responses Reveal Negative Feedback by BHLHE40 and BHLHE41. Cell Reports 33, 108371 (2020). https://doi.org/10.1016/j.celrep.2020.108371

Back to the plate

2020, Scientific Reports. Well C7.

On the correlation between material-induced cell shape and phenotypical response of human mesenchymal stem cells

Read the paper

Caption

Clustering of cell shape diversity. (a) Surface topography as a model system to investigate cell shape-induced cells response. TCP: tissue culture polystyrene. The schematic was prepared in the Inkscape 0.92.4. (b) The montage image that represents the diverse cell of hMSCs seeded on different topographies and flat surface (upper, left). Nuclei were stained with DAPI (magenta), actin was stained with phalloidin (cyan). The collage was prepared in the GIMP v 2.10.14. (c) Principal components analysis of all cell shapes on the TopoChip. The first two principal components are shown. Note that no noticeable clusters are observed. (d) Cluster gram representing a clustering of TopoChip cell shape data. Blue and yellow colours distinguish different clusters. (e) Heatmap representing the median values of cell shape features on the chosen medoid topographies. (f) tSNE (t-distributed stochastic neighbor embedding) visualization of the obtained clusters. Cells shape silhouettes on 28 selected surfaces are represented.

Citation

Aliaksei S. Vasilevich, Steven Vermeulen, Marloes Kamphuis, Nadia Roumans, Said Eroumé, Dennie G. A. J. Hebels, Jeroen van de Peppel, Rika Reihs, Nick R. M. Beijer, Aurélie Carlier, Anne E. Carpenter, Shantanu Singh, Jan de Boer. On the correlation between material-induced cell shape and phenotypical response of human mesenchymal stem cells. Scientific Reports 10, 18988 (2020). https://doi.org/10.1038/s41598-020-76019-z

Back to the plate

2020, Scientific Reports. Well C8.

Evolutionary design of optimal surface topographies for biomaterials

Aliaksei Vasilevich, Aurélie Carlier, David A. Winkler, Shantanu Singh, Jan de Boer

Read the paper

Caption

Selecting, recombining and mutating the genetic information of the parents. (A). Schematic representation of surface selection flow at the beginning of each cycle. (B). Schematic representation of crossover and mutation algorithms. (C) Example of different crossover algorithms. Colours (yellow and blue) represent parts from different parents. Surfaces with red outline were selected further to demonstrate mutation algorithms. (D) Example of different mutation algorithms. Green regions correspond to newly-inserted elements of the design; white outlines corresponds to deleted parts.

Citation

Aliaksei Vasilevich, Aurélie Carlier, David A. Winkler, Shantanu Singh, Jan de Boer. Evolutionary design of optimal surface topographies for biomaterials. Scientific Reports 10, 22160 (2020). https://doi.org/10.1038/s41598-020-78777-2

Back to the plate

2021

2021, Molecular Biology of the Cell. Well C9.

Predicting cell health phenotypes using image-based morphology profiling

Read the paper

Caption

Data processing and modeling approach. (a) Example images and workflow from the Cell Health assays. We apply a series of manual gating strategies (see Materials and Methods) to isolate cell subpopulations and to generate cell health readouts for each perturbation. (Top) In the “Cell Cycle” panel, in each nucleus we measure Hoechst, EdU, PH3, and gH2AX. (Bottom) In the “Viability” panel, we capture DPC images, measure Caspase 3/7, DRAQ7, and CellROX. (b) Example Cell Painting image across five channels, plus a merged representation across channels. The image is cropped from a larger image and shows ES2 cells. Scale bars are 20 µm. Below are the steps applied in an image-based profiling pipeline, after features have been extracted from each cell’s image. (c) Modeling approach where we fit 70 different regression models using CellProfiler features derived from Cell Painting images to predict Cell Health readouts. Model weights refer to the coefficients derived from each regression model.

Citation

Gregory P. Way, Maria Kost-Alimova, Tsukasa Shibue, William F. Harrington, Stanley Gill, Federica Piccioni, Tim Becker, Hamdah Shafqat-Abbasi, William C. Hahn, Anne E. Carpenter, Francisca Vazquez, Shantanu Singh. Predicting cell health phenotypes using image-based morphology profiling. Molecular Biology of the Cell 32, 995-1005 (2021). https://doi.org/10.1091/mbc.E20-12-0784

Back to the plate

2021, eNeuro. Well C10.

Molecular Diversity of Glutamatergic and GABAergic Synapses from Multiplexed Fluorescence Imaging

Eric Danielson, Karen Perez de Arce, Beth Cimini, Eike-Christian Wamhoff, Shantanu Singh, Jeffrey R. Cottrell, Anne E. Carpenter, Mark Bathe

Read the paper

Caption

Synaptic cluster identification using UMAP. A, UMAP plots of individual synapses (n = 10,000) using CellProfiler output separates excitatory (red) and inhibitory (blue) synapses into two major clusters with multiple subclusters. B, Unique clusters identified by HDBSCAN. (–) indicates synaptic target below limit of detection. C, Representative synapses from each cluster. D, Heatmap indicates the average relative intensities for each synaptic target within each cluster. All values are normalized to untreated mean integrated intensity.

Citation

Eric Danielson, Karen Perez de Arce, Beth Cimini, Eike-Christian Wamhoff, Shantanu Singh, Jeffrey R. Cottrell, Anne E. Carpenter, Mark Bathe. Molecular Diversity of Glutamatergic and GABAergic Synapses from Multiplexed Fluorescence Imaging. eNeuro 8, ENEURO.0286-20.2020 (2021). https://doi.org/10.1523/ENEURO.0286-20.2020

Back to the plate

2022

2022, PLOS Computational Biology. Well C11.

Predicting drug polypharmacology from cell morphology readouts using variational autoencoder latent space arithmetic

Yuen Ler Chow, Shantanu Singh, Anne E. Carpenter, Gregory P. Way

Read the paper

Caption

Our variational autoencoder (VAE) implementation framework, applied to determining the phenotype of cells. One application is to predict the phenotype of cells treated with compounds that have two mechanisms of actions (MOA), given the phenotype of cells treated with compounds that have each of those single MOAs (bottom right). A VAE encodes input data into a lower-dimensional latent space and then decodes the representation back into the original data dimensions. Our data contained measurements for 588 morphology features, each averaged for each population of cells treated with a given chemical compound. Following a sweep to select optimal hyperparameters (see Methods), we set our latent space dimension to 10 dimensions. The vanilla VAE learns by minimizing a reconstruction and KL-divergence loss. The other VAE variants we tested minimize loss functions that encourage disentangled features that promote interpretability, data simulation, and enable meaningful LSA.

Citation

Yuen Ler Chow, Shantanu Singh, Anne E. Carpenter, Gregory P. Way. Predicting drug polypharmacology from cell morphology readouts using variational autoencoder latent space arithmetic. PLOS Computational Biology 18, 1009888 (2022). https://doi.org/10.1371/journal.pcbi.1009888

Back to the plate

2022, Molecular Biology of the Cell. Well C12.

Cell Painting predicts impact of lung cancer variants

Juan C. Caicedo, John Arevalo, Federica Piccioni, Mark-Anthony Bray, Cathy L. Hartland, Xiaoyun Wu, Angela N. Brooks, Alice H. Berger, Jesse S. Boehm, Anne E. Carpenter, Shantanu Singh

Read the paper

Caption

Morphology-based variant impact phenotyping (cmVIP) and resulting predictions in a diverse set of variants. (A) Decision tree of the VIP algorithm (Berger et al., 2017; Thornton et al., 2021), which we adopt for classifying variants by their Cell Painting profiles as gain of function (GOF), loss of function (LOF), change of function (COF), and neutral (NT) mutations. (B) Example predictions by cmVIP on four variants, one of each type. The correlation matrices at the top show how similar the replicates of each pair are (reference gene self-correlation, reference–variant cross-correlation, and variant self-correlation). The correlation matrix colors represent the correlation values in the same color scale as in D. The boxplots below the matrices show the distribution of median values of the matrices’ rows (self-correlation) and columns (cross-correlation). (C) Sparkler plots display the magnitude and directionality of predictions for all variants in a gene set. The x-axis represents the negative log p-value of the impact test (the larger the more impactful), and the y-axis represents the log p-value of the directionality test polarized by the result of the strength test. All variants for these genes are displayed, but only a few are annotated to aid visualization. All the plots and annotations can be queried at full scale in the interactive website: http://broad.io/cmvip. (D) Correlation matrices for the groups of variants presented in C, together with the corresponding reference alleles.

Citation

Juan C. Caicedo, John Arevalo, Federica Piccioni, Mark-Anthony Bray, Cathy L. Hartland, Xiaoyun Wu, Angela N. Brooks, Alice H. Berger, Jesse S. Boehm, Anne E. Carpenter, Shantanu Singh. Cell Painting predicts impact of lung cancer variants. Molecular Biology of the Cell 33, ar49 (2022). https://doi.org/10.1091/mbc.E21-11-0538

Back to the plate

Figure 1.

2022, Cell Systems. Well D1.

Virtual screening for small-molecule pathway regulators by image-profile matching

Mohammad H. Rohban, Ashley M. Fuller, Ceryl Tan, Jonathan T. Goldstein, Deepsing Syangtan, Amos Gutnick, Ann DeVine, Madhura P. Nijsure, Megan Rigby, Joshua R. Sacher, Steven M. Corsello, Grace B. Peppler, Marta Bogaczynska, Andrew Boghossian, Gabrielle E. Ciotti, Allison T. Hands, Aroonroj Mekareeya, Minh Doan, Jennifer P. Gale, Rik Derynck, Thomas Turbyville, Joel D. Boerckel, Shantanu Singh, Laura L. Kiessling, Thomas L. Schwarz, Xaralabos Varelas, Florence F. Wagner, Ran Kafri, T.S. Karin Eisinger-Mathason, Anne E. Carpenter

Read the paper

Caption

Image profile-based drug discovery offers efficient, virtual screening for pathway modulators. a) If an overexpressed gene changes the morphology of cells, its image-based profile can be used as a query in a database of small molecule profiles, looking for those that match (positively correlate) or oppose (negatively correlate). b) Cell Painting images for two positive control gene-compound matches that yield observable morphological phenotypes (not all are expected to). EMPTY and DMSO are the negative controls in the gene overexpression and compound experiments, respectively; they differ in confluency and image acquisition conditions. The phenotype of p38α (MAPK14) overexpression matches (correlates to) that of SB-203580, a known p38 inhibitor; in both, elongated cells are over-represented. The phenotype of CDK2 overexpression (small cells) negatively correlates to that of purvalanol-a, a known CDK inhibitor, which induces an opposite phenotype (huge cells). Scale bars= 60 μm. c) Enrichment plot of all gene-compound pairs sorted based on their absolute profile correlation. Starting from the left, the curve rises a unit if the gene-compound pair is a known/annotated connection, and goes down a unit otherwise. The units are normalized to the number of possible known connections, so the maximum height is one and ends in zero. The steep increase of the curve indicates enrichment of correct connections towards the top of the rank-ordered list of pairs.

Citation

Mohammad H. Rohban, Ashley M. Fuller, Ceryl Tan, Jonathan T. Goldstein, Deepsing Syangtan, Amos Gutnick, Ann DeVine, Madhura P. Nijsure, Megan Rigby, Joshua R. Sacher, Steven M. Corsello, Grace B. Peppler, Marta Bogaczynska, Andrew Boghossian, Gabrielle E. Ciotti, Allison T. Hands, Aroonroj Mekareeya, Minh Doan, Jennifer P. Gale, Rik Derynck, Thomas Turbyville, Joel D. Boerckel, Shantanu Singh, Laura L. Kiessling, Thomas L. Schwarz, Xaralabos Varelas, Florence F. Wagner, Ran Kafri, T. S. Karin Eisinger-Mathason, Anne E. Carpenter. Virtual screening for small-molecule pathway regulators by image-profile matching. Cell Systems 13, 724-736.e9 (2022). https://doi.org/10.1016/j.cels.2022.08.003

Back to the plate

2022, Frontiers in Pharmacology. Well D2.

In vitro neutrophil migration is associated with inhaled corticosteroid treatment and serum cytokines in pediatric asthma

Solveig Lemmel, Markus Weckmann, Anna Wohlers, Adan Chari Jirmo, Ruth Grychtol, Isabell Ricklefs, Gyde Nissen, Anna Bachmann, Shantanu Singh, Juan Caicedo, Thomas Bahmer, Gesine Hansen, Erika Von Mutius, Klaus F. Rabe, Oliver Fuchs, Anna-Maria Dittrich, Bianca Schaub, Christine Happle, Anne E. Carpenter, Matthias Volkmar Kopp, Tim Becker, the ALLIANCE Study Group as part of the German Centre for Lung Research (DZL)

Read the paper

Caption

Pipeline of the SiMA chemotaxis analysis. (A) Neutrophils are isolated from a blood sample and chemotaxis is induced using a SiMA assay. For each patient, the migration towards IL8, fMLP and LTB₄ is recorded using time lapse microscopy. (B) A Deep Learning based image segmentation (UNet model) is used to identify and track neutrophil granulocytes. For each experiment, single-cell and single-trajectory profiles are created using CellProfiler and MigrationmineR, describing each neutrophil with a total of 25 features (after feature selection). These features are aggregated on an experiment level and used for quantification of chemotaxis. (C) The most telling and interpretable migration parameters are speed, chemotactic response and the gradient precision. A multidimensional analysis is performed using the complete feature set, which is projected into a two dimensional SiMA landscape using the UMAP algorithm for dimension reduction.

Citation

Solveig Lemmel, Markus Weckmann, Anna Wohlers, Adan Chari Jirmo, Ruth Grychtol, Isabell Ricklefs, Gyde Nissen, Anna Bachmann, Shantanu Singh, Juan Caicedo, Thomas Bahmer, Gesine Hansen, Erika Von Mutius, Klaus F. Rabe, Oliver Fuchs, Anna-Maria Dittrich, Bianca Schaub, Christine Happle, Anne E. Carpenter, Matthias Volkmar Kopp, Tim Becker, the ALLIANCE Study Group as part of the German Centre for Lung Research (DZL). In vitro neutrophil migration is associated with inhaled corticosteroid treatment and serum cytokines in pediatric asthma. Frontiers in Pharmacology 13, 1021317 (2022). https://doi.org/10.3389/fphar.2022.1021317

Back to the plate

2022, Cell Systems. Well D3.

Morphology and gene expression profiling provide complementary information for mapping cell state

Read the paper

Caption

Cell morphology predicts the bortezomib sensitivity of clones across datasets. Box plots of Bortezomib Signatures for clones in the (A) training, (B) validation, (C) test, and (D) holdout datasets. Plots show values for individual well profiles (points), range (error bars), 25th and 75th percentiles (box boundaries), and median. Dashed vertical black line is Bortezomib Signature = 0, dashed vertical red lines are the 95% confidence interval for Bortezomib Signatures of 1000 random permutations of the data. (E) ROC curves for the performance of the Bortezomib Signature on the indicated dataset (solid line) or its shuffled counterpart (dashed line). Datasets are designated by color: training (magenta), validation (orange), test (teal), and holdout (purple). Colored points are the corresponding false positive and true positive rates at the absolute minimum thresholds for each respective dataset. Black dotted line is the identity line where false positive rate = true positive rate. AUROC values reported for data and shuffled data. See Figure 3—source data 1 for breakdown of profiles and experiments per dataset.

Citation

Megan E. Kelley, Adi Y. Berman, David R. Stirling, Beth A. Cimini, Yu Han, Shantanu Singh, Anne E. Carpenter, Tarun M. Kapoor, Gregory P. Way. High-content microscopy reveals a morphological signature of bortezomib resistance. eLife 12, e91362 (2023). https://doi.org/10.7554/eLife.91362

Back to the plate

2024

2024, Advances in Neural Information Processing Systems (NeurIPS 2024). Well E3.

MOTIVE: A Drug-Target Interaction Graph For Inductive Link Prediction

John Arevalo, Ellen Su, Anne E. Carpenter, Shantanu Singh

Read the paper

Caption

Schematic of the random split (top row) and cold-source split (bottom row). The leftmost graph illustrates the actual partitioning of the edges, and the three graphs to the right show which nodes and edges are visible to the GNN models during training, validation, and testing. The number of edges in each partition is not representative of our true 70/10/20 ratios. Cold-target split is symmetrical to cold-source split. The model aggregates neighbor features via the message-passing edges (solid lines) and makes predictions on the supervision edges (dotted lines).

Citation

John Arevalo, Ellen Su, Anne E. Carpenter, Shantanu Singh. MOTIVE: A Drug-Target Interaction Graph For Inductive Link Prediction. Advances in Neural Information Processing Systems (NeurIPS 2024) 37, 140320-140333 (2024). https://doi.org/10.52202/079017-4454

Back to the plate

2024, Chapter 4 of F. Garcia Fóssa, PhD thesis, Universidade Estadual de Campinas. Well E4.

Image Profiling of Huh-7 Hepatocytes Reveals Distinctive Silver Nanoparticle Effects Over 15 Days

Fernanda Garcia-Fossa, Thais Lacerda de Moraes, Shantanu Singh, Anne E. Carpenter, Beth A. Cimini, Marcelo Bispo de Jesus

Caption

UMAP plots of Cell Painting, Live Cell Painting, and Tox Painting panels of Huh-7 cells exposed to AgNP for 1 and 15 days. UMAP representations of each panel labeled by the exposure time (1 or 15 days) in the left panel, and their respective K-means clustering-labeled UMAPs in the right panel, in 2D and 3D. Each point represents the median aggregated data per well in the experiment. UMAPs ran for 50 iterations, so the mean embedding and the standard deviation are represented. The larger dots represent the cells treated with AgNP 100 nm, intermediate points AgNP 40 nm, and the smallest are the negative controls. This assay was performed using n = 5 replicates and repeated twice independently.

Citation

Fernanda Garcia-Fossa, Thais Lacerda de Moraes, Shantanu Singh, Anne E. Carpenter, Beth A. Cimini, Marcelo Bispo de Jesus. Image Profiling of Huh-7 Hepatocytes Reveals Distinctive Silver Nanoparticle Effects Over 15 Days. Chapter 4 of F. Garcia Fóssa, PhD thesis, Universidade Estadual de Campinas, 108-144 (2024).

Back to the plate

2024, Nature Cell Biology. Well E5.

Bringing computation to biology by bridging the last mile

Anne E. Carpenter, Shantanu Singh

Read the paper

Caption

Stages of technology-driven acceleration in biological research that involves computational methods.

Citation

Anne E. Carpenter, Shantanu Singh. Bringing computation to biology by bridging the last mile. Nature Cell Biology 26, 5-7 (2024). https://doi.org/10.1038/s41556-023-01286-7

Back to the plate

2024, Nature Communications. Well E6.

High-dimensional phenotyping to define the genetic basis of cellular morphology

Read the paper

Caption

Study overview. iPSC lines from 297 donors were expanded, quality-control checked and then subject to both high-throughput imaging with Cell Painting and 30X whole-genome sequencing (WGS). Overall, we imaged 5.1 × 10⁶ cells across all donors and quantified 3418 morphological traits per cell using CellProfiler software. We inferred genetic variants from the WGS data and investigated whether individual morphological traits associated with both rare and common variation.

Citation

Matthew Tegtmeyer, Jatin Arora, Samira Asgari, Beth A. Cimini, Ajay Nadig, Emily Peirent, Dhara Liyanage, Gregory P. Way, Erin Weisbart, Aparna Nathan, Tiffany Amariuta, Kevin Eggan, Marzieh Haghighi, Steven A. McCarroll, Luke O’Connor, Anne E. Carpenter, Shantanu Singh, Ralda Nehme, Soumya Raychaudhuri. High-dimensional phenotyping to define the genetic basis of cellular morphology. Nature Communications 15, 347 (2024). https://doi.org/10.1038/s41467-023-44045-w

Back to the plate

Figure 5 from Insights into Drug Cardiotoxicity from Biological and Chemical Data: The First Public Classifiers for FDA Drug-Induced Cardiotoxicity Rank — Figure 5.

2024, Journal of Chemical Information and Modeling. Well E7.

Insights into Drug Cardiotoxicity from Biological and Chemical Data: The First Public Classifiers for FDA Drug-Induced Cardiotoxicity Rank

Srijit Seal, Ola Spjuth, Layla Hosseini-Gerami, Miguel García-Ortegón, Shantanu Singh, Andreas Bender, Anne E. Carpenter

Read the paper

Caption

Comparison of evaluation metrics of models built in this study with an external test set of 90 compounds evaluated by the (a) AUC-ROC and (b) AUCPR and (c) performance of each model across compounds that are similar (dissimilar) to the training data. The ensemble model in (c) is based on the models built on six data sources listed in the main text, and evaluation is for 90 held-out compounds; results for the ensemble using 11 data sources and 78 held-out compounds are in Supporting Information Figure S3.

Citation

Srijit Seal, Ola Spjuth, Layla Hosseini-Gerami, Miguel García-Ortegón, Shantanu Singh, Andreas Bender, Anne E. Carpenter. Insights into Drug Cardiotoxicity from Biological and Chemical Data: The First Public Classifiers for FDA Drug-Induced Cardiotoxicity Rank. Journal of Chemical Information and Modeling 64, 1172-1186 (2024). https://doi.org/10.1021/acs.jcim.3c01834

Back to the plate

2024, Nature Communications. Well E8.

Learning representations for image-based profiling of perturbations

Read the paper

Caption

Quantitative and qualitative evaluation of feature representations of treatment effects. The evaluation task is biological profile matching (see Fig. 1G). A Performance of feature representations for the three benchmark datasets according to two metrics: Mean Average Precision (MAP) in the x axis and Folds of Enrichment in the y axis (see Methods). Each point indicates the mean of these metrics over all queries using the following feature representations: CellProfiler (pink), a CNN pre-trained on ImageNet (yellow), a CNN trained on the combined set of Cell Painting images (cyan), and a CNN trained on Cell Painting images from the same dataset (green). In all cases, sphering batch-correction was applied on well-level profiles. B 2D UMAP projections of treatment profiles obtained with the Cell Painting CNN (672 features) after batch correction for the three datasets evaluated in this work. The plot includes a projection of well-level profiles (gray points), control wells (red points), and aggregated treatment-level profiles of treatments (blue points). Dashed lines indicate clusters of treatment-level profiles where all or the majority of points share the same biological annotation. Source data is provided as a Source Data file.

Citation

Nikita Moshkov, Michael Bornholdt, Santiago Benoit, Matthew Smith, Claire McQuin, Allen Goodman, Rebecca A. Senft, Yu Han, Mehrtash Babadi, Peter Horvath, Beth A. Cimini, Anne E. Carpenter, Shantanu Singh, Juan C. Caicedo. Learning representations for image-based profiling of perturbations. Nature Communications 15, 1594 (2024). https://doi.org/10.1038/s41467-024-45999-1

Back to the plate

Figure 1 from From pixels to phenotypes: Integrating image-based profiling with cell health data as BioMorph features improves interpretability — Figure 1.

2024, Molecular Biology of the Cell. Well E9.

From pixels to phenotypes: Integrating image-based profiling with cell health data as BioMorph features improves interpretability

Srijit Seal, Jordi Carreras-Puigvert, Shantanu Singh, Anne E. Carpenter, Ola Spjuth, Andreas Bender

Read the paper

Caption

A map of the BioMorph space. A general representation of the hierarchy of levels is shown (with examples) for each BioMorph term that is organised from Cell Painting features (level 5) and containing information on Cell Health (level 3) associated with measurement type (level 2) under an assay type (level 1) associated with Cell process affected (level 4). Further terms in Supplemental Figure S1 with all terms listed in Supplemental Table S1.

Citation

Srijit Seal, Jordi Carreras-Puigvert, Shantanu Singh, Anne E. Carpenter, Ola Spjuth, Andreas Bender. From pixels to phenotypes: Integrating image-based profiling with cell health data as BioMorph features improves interpretability. Molecular Biology of the Cell 35, mr2 (2024). https://doi.org/10.1091/mbc.E23-08-0298

Back to the plate

2024, Nature Methods. Well E10.

Robert van Dijk, John Arevalo, Mehrtash Babadi, Anne E. Carpenter, Shantanu Singh

Read the paper

Caption

(a) Human U2OS cells treated with dimethyl sulfoxide (DMSO) and stained using the Cell Painting assay, which employs six dyes in five channels to label eight cellular compartments. The top row (from left to right) shows mitochondrial staining; actin, Golgi, and plasma membrane staining; and nucleolar and cytoplasmic RNA staining. The bottom row (from left to right) displays endoplasmic reticulum staining, DNA staining, and a montage of all five channels (from Cimini et al. [21]). (b) Thousands of features are extracted from each segmented cell in microscopy images of wells. A learned function f(x) (CytoSummaryNet) aggregates this data into a single feature vector: the sample’s profile. (c) An in-depth look at the model architecture used in this study. The model consists of three elements: a function φ(x), which maps the input data from ℝ^D to ℝ^L space, a summation, which collapses the cell dimension, and ρ(z), which maps the collapsed representation from ℝ^N to ℝ^L space. (d) During training, replicate compound profiles are forced to attract each other (green arrows) and simultaneously repel every other compound (red arrows) in the learned feature space. Here, all forces are drawn for a single profile of compound B.

Citation

Robert van Dijk, John Arevalo, Mehrtash Babadi, Anne E. Carpenter, Shantanu Singh. Capturing cell heterogeneity in representations of cell populations for image-based profiling using contrastive learning. PLOS Computational Biology 20, 1012547 (2024). https://doi.org/10.1371/journal.pcbi.1012547

Back to the plate

2025

2025, arXiv. Well F7.

Erik Serrano, Srinivas Niranj Chandrasekaran, Dave Bunten, Kenneth I. Brewer, Jenna Tomkinson, Roshan Kern, Michael Bornholdt, Stephen J. Fleming, Ruifan Pei, John Arevalo, Hillary Tsang, Vincent Rubinetti, Callum Tromans-Coia, Tim Becker, Erin Weisbart, Charlotte Bunne, Alexandr A. Kalinin, Rebecca Senft, Stephen J. Taylor, Nasim Jamali, Adeniyi Adeboye, Hamdah Shafqat Abbasi, Allen Goodman, Juan C. Caicedo, Anne E. Carpenter, Beth A. Cimini, Shantanu Singh, Gregory P. Way

Read the paper

Caption

The standard image-based profiling experiment and the role of Pycytominer. a, In the experimental phase, a scientist plates cells, often perturbing them with chemical or genetic agents, and performs microscopy imaging. In image analysis, using CellProfiler for example, a scientist applies several data-processing steps to generate image-based profiles. In addition, scientists can apply deep-learning tools, such as DeepProfiler, to generate image-based profiles. b, Pycytominer performs image-based profiling to process morphology features and make them ready for downstream analyses. c, Pycytominer performs five fundamental functions, each implemented with a simple and intuitive API. Each function enables a user to implement various methods for executing operations. Created in BioRender.com.

Citation

Back to the plate

2025, bioRxiv. Well G1.

Cell morphological representations of genes enhance prediction of drug targets

Srinivas Niranj Chandrasekaran, Eric Alix, John Arevalo, Adriana Borowa, Patrick J. Byrne, William G. Charles, Zitong S. Chen, Beth A. Cimini, Boxiong Deng, John G. Doench, Jessica D. Ewald, Briana Fritchman, Colin J. Fuller, Jedidiah Gaetz, Amy Goodale, Marzieh Haghighi, Yu Han, Zahra Hanifehlou, Holger Hennig, Desiree Hernandez, Christina B. Jacob, Tim James, Tomasz Jetka, Alexandr A. Kalinin, Ben Komalo, Maria Kost-Alimova, Tomasz Krawiec, Brittany A. Marion, Glynn Martin, Nicola Jane McCarthy, Lisa Miller, Arne Monsees, Nikita Moshkov, Alán F. Muñoz, Arnaud Ogier, Magdalena Otrocka, Krzysztof Rataj, David E. Root, Francesco Rubbo, Simon Scrace, Douglas W. Selinger, Rebecca A. Senft, Peter Sommer, Amandine Thibaudeau, Sarah Trisorus, Rahul Valiya Veettil, William J. Van Trump, Sui Wang, Michał Warchoł, Erin Weisbart, Amélie Weiss, Michael Wiest, Agata Zaremba, Andrei Zinovyev, Shantanu Singh, Anne E. Carpenter

Read the paper

Morphological profiling using Cell Painting. a, Schematic of the Cell Painting assay; cells are incubated and perturbed and a set of six stains is applied. b, Images are then obtained in five channels by automated microscopy followed by nucleus and cell body segmentation. c, Appropriate software or deep-learning-based methods are applied to measure or calculate morphological features from the images. d, After feature preprocessing, downstream analysis is performed. This includes a variety of methods, including supervised and unsupervised machine learning, to better elucidate the biological effects of a compound, such as its mechanism of action or safety profile. e,f, Adaptations of the Cell Painting assay include BODIPY to mark lipid droplets in lipid-accumulating cells (e) and a coronavirus antibody against human coronavirus 229E (CoV-229E) viral protein (f).

Citation

Srijit Seal, Maria-Anna Trapotsi, Ola Spjuth, Shantanu Singh, Jordi Carreras-Puigvert, Nigel Greene, Andreas Bender, Anne E. Carpenter. Cell Painting: a decade of discovery and innovation in cellular imaging. Nature Methods 22, 254-268 (2025). https://doi.org/10.1038/s41592-024-02528-8

Back to the plate

2026

2026, arXiv. Well H1.

JUMP-lite: Compact, reproducible benchmarking of cell representations

Alán F. Muñoz, Johan Fredin Haslum, Runxi Shen, Anne E. Carpenter, Shantanu Singh

Read the paper

Caption

JUMP-lite substantially reduces storage while focusing on annotated perturbations for benchmarking. (a) Compound candidates were mapped to the union of MOTIVE’s compound–compound (CC) and compound–gene (CG) graphs, then passed through filters for plate quality, profile quality and replicate count. Supplementary Section S1.1 provides detailed component and annotation accounting, and Supplementary Figure S1 shows target overlap across the compound, CRISPR, and ORF components. (b) Disk-size comparison of JUMP Cell Painting with lossless and lossy compression options. Each square’s area is directly proportional to its disk size. (c) Standardized pipeline from images to evaluation. ALIBY [29] orchestrates segmentation (Cellpose) and feature extraction (cp measure), while Nahual connects isolated environments for deep learning models. A normalization sweep processes the profiles for comparison across methods.

Citation

Alán F. Muñoz, Johan Fredin Haslum, Runxi Shen, Anne E. Carpenter, Shantanu Singh. JUMP-lite: Compact, reproducible benchmarking of cell representations. arXiv (2026). https://doi.org/10.48550/arXiv.2608.07632

Back to the plate

2026, arXiv. Well H2.

Science sandboxes measure the scientific capability of AI agents

Arya S. Rao, Rodrigo I. Castro, Sager J. Gosai, Kenneth B. Hsu, Yasha Ektefaie, Shantanu Singh, Sangeeta N. Bhatia, Steven K. Reilly, Ryan Tewhey, Eric S. Lander, Pardis C. Sabeti

Read the paper

With the blog post

Measuring the frontier of scientific capability in the age of AI

Arya Rao, Bryan Hsu, Yasha Ektefaie, Jackson Weir, Sandeep Kambhampati, Frank Liu, Fei Chen, Sangeeta Bhatia, Shantanu Singh, Eric Lander, Pardis Sabeti

Read the blog post

Caption

Science sandboxes for closed-loop evaluation of AI experimentalists. (a) A science sandbox couples an autonomous agent to a sealed oracle through a closed loop of action, feedback, and hypothesis revision. At each design round, the agent proposes a batch of actions, receives feedback from the oracle, records its interpretation, and selects its next actions. The oracle can be wet (a physical experiment), damp (a model trained on experimental data), or dry (an invented rule created by the experimenter). (b) MPRAbox instantiates this framework in regulatory genomics. The agent selects a finite massively parallel reporter assay (MPRA) library from the effectively unbounded space of possible 200 bp regulatory DNA sequences. The library is evaluated by a sealed in silico MPRA oracle that acts as a simulated wet lab, and aggregate performance metrics from diverse hidden evaluation sets are returned to guide subsequent design rounds.

Citation

Arya S. Rao, Rodrigo I. Castro, Sager J. Gosai, Kenneth B. Hsu, Yasha Ektefaie, Shantanu Singh, Sangeeta N. Bhatia, Steven K. Reilly, Ryan Tewhey, Eric S. Lander, Pardis C. Sabeti. Science sandboxes measure the scientific capability of AI agents. arXiv (2026). https://doi.org/10.48550/arXiv.2608.30165

Arya Rao, Bryan Hsu, Yasha Ektefaie, Jackson Weir, Sandeep Kambhampati, Frank Liu, Fei Chen, Sangeeta Bhatia, Shantanu Singh, Eric Lander, Pardis Sabeti. Measuring the frontier of scientific capability in the age of AI. Blog post (2026). https://aryarao.substack.com/p/measuring-the-frontier-of-scientific

Back to the plate

2026, arXiv. Well H3.

Same Encoder, Different Winner: A Paired-View Framework for Cell Painting Encoder Evaluation

Tim Treis, Nikita Moshkov, Johan Fredin Haslum, Shantanu Singh, Fabian J. Theis

Read the paper

Caption

Overview of the CP-BG-Bench framework. Left: data engineering pipeline. Source-specific adapters ingest JUMP-CP, RxRx1, and RxRx3-core; images are preprocessed, single-cell crops are extracted, and per-cell quality control yields the final cell pool. Center: four matched views are derived from each cell tile while preserving identity. The raw crop (C); segmented (S) zeroes out everything outside the cell mask; crops with density patches (CD) and segmented with density patches (SD) overlay four corner patches whose intensity encodes local cell count, providing an explicit, source-agnostic density signal. Right: each of the twelve resulting encoder-view configurations per dataset (3 encoders × 4 views) is fine-tuned with LoRA against perturbation embeddings under a multi-positive contrastive objective, producing 36 trained checkpoints in total. The resulting embeddings are evaluated under four community-standard protocols (replicate mAP, scIB batch integration, CellProfiler feature prediction, cross-batch perturbation recall).

Citation

Tim Treis, Nikita Moshkov, Johan Fredin Haslum, Shantanu Singh, Fabian J. Theis. Same Encoder, Different Winner: A Paired-View Framework for Cell Painting Encoder Evaluation. arXiv (2026). https://doi.org/10.48550/arXiv.2609.12761

Back to the plate

2026, Nature Communications. Well H4.

Read the paper

Caption

The JUMP consortium generated Cell Painting data for 115,796 compounds in cpg0016-jump. (A) The JUMP Cell Painting dataset includes cpg0000, cpg0001, cpg0002, and cpg0016 (Methods). This study presents the 115,796 chemical compounds of cpg0016, generated across the 12 partner sources/laboratories (note: Source 7 and 13 correspond to the same laboratory). Typically, five replicates of each compound were generated at two to four laboratories besides the one that nominated the compound. The arrows denote the compound exchange logistics (Supplementary Table 1) and apply only to the ten wave 1/wave 2 nominated-compound sources; S4, S7, and S13 did not participate in that exchange. cpg0016 genetic perturbations were previously published ²¹: CRISPR data generated entirely at S13 and ORFs at S4, the Broad. The size and overlaps in the schematic are not precise to scale. Overlap is also shown for cpg0012 ²⁰, Broad’s existing public dataset of 30,616 bioactive compounds, reprocessed using the JUMP processing pipeline (Methods). (B) Example Cell Painting images. Top: One negative control field of view (DMSO, dimethyl sulfoxide) showing the five imaged channels, pseudocolored, plus their merge. Bottom: visually strong phenotypes in each channel for various compounds. Single-channel and merged images were individually scaled to the 99th intensity percentiles and full intensity range of the image for visibility, respectively. All images are from a single source (S5) for consistency. Scale bars = 25 μm. (C-F) Chemical characterization of the 112,462 analyzed JUMP compounds: (C) Median physicochemical properties: QED (Quantitative Estimate of Drug-likeness ²⁴); MW (molecular weight); LogP (calculated octanol-water partition coefficient); TPSA (topological polar surface area); HBD (hydrogen-bond donors); HBA (hydrogen-bond acceptors). (D) Lipinski rule-of-five (Ro5) compliance. (E) Clinical development stage for the 1,731 compounds with such annotation. (F) Number and percentage of compounds with a target or mechanism of action (MOA) annotation from the Drug Repurposing Hub, ChEMBL, the Chemical Probes Portal, and MOTIVE.

Citation

Back to the plate

2026, Manuscript and computational supplement, version 1.0.2, Broad Institute of MIT and Harvard. Well H11.

Average precision under random ranking: baseline, variability, and formal verification

Shantanu Singh

Read the paper

Caption

Exact moments and a discrete AP law. (A) Mean divided by prevalence for M/L = 0.1 and for M = 3. (B) Variance multiplied by L; dotted lines mark the respective limits 0.09 and π²/18. (C) All 28 label vectors with L = 8, M = 2 produce 26 distinct AP values; equal values are combined into probability masses. The dashed line is the mean, and the shaded band is one standard deviation on either side.

Citation

Shantanu Singh. Average precision under random ranking: baseline, variability, and formal verification. Manuscript and computational supplement, version 1.0.2, Broad Institute of MIT and Harvard (2026). https://github.com/shntnu/expected_ap/releases/tag/v1.0.2

Back to the plate

2026, Essay. Well H12.

Propel

Jackson Weir, Matteo Di Bernardo, Yasha Ektefaie, Arya Rao, Luca Pinello, Shantanu Singh

Read the essay

Caption

The essay's six line drawings, one for each of the first questions, clockwise from the top: spatial genomics, liver toxicity, regulatory DNA across cell types, optical pooled screening, rapid experimental loops, and human adaptation. The drawings are from the essay, and the ring was arranged for this tribute.

Citation

Jackson Weir, Matteo Di Bernardo, Yasha Ektefaie, Arya Rao, Luca Pinello, Shantanu Singh. Propel. Essay (2026). https://shntnu.bio/propel/

Back to the plate