Daniel R. Wong
Research Subjects
Daniel R. Wong has a total of 58 co-authors. Here are the top co-authors by number of shared publications:
- Michael J. Keiser
Institute for Neurodegenerative Diseases, University of California, San Francisco, San Francisco, USA
5 shared publications - Brittany N. Dugger
University of California Davis Health, Department of Pathology and Laboratory Medicine, CA, USA
4 shared publications - Atul Janardhan Butte
Bakar Computational Health Sciences Institute, University of California, San Francisco, San Francisco, USA
3 shared publications Andrew Kiruluta
Machine Learning and Computational Sciences, Pfizer Worldwide Research Development and Medical, 610 Main Street, Cambridge, Massachusetts 02139, USA
2 shared publications- Charles DeCarli
Department of Neurology, University of California, Davis
2 shared publications - Charles L. White
University of Texas Southwestern Medical Center
2 shared publications
Latest Publications
- Open Access
- 11 November 2026
- Published by Royal Society of Chemistry (RSC) in Digital Discovery
- Vol. 5  (11) , 4924-4937
Phenotypic reversion and target prioritization for cellular inflammation via representation learning with foundation models
The identification of genetic perturbations that can reverse disease-associated cellular phenotypes toward a healthy state is a central challenge in early drug discovery.
- 11 November 2026
- Published by Royal Society of Chemistry (RSC) in Digital Discovery
- Vol. 5  (11) , 4924-4937
Open AccessPhenotypic reversion and target prioritization for cellular inflammation via representation learning with foundation models
The identification of genetic perturbations that can reverse disease-associated cellular phenotypes toward a healthy state is a central challenge in early drug discovery.
- Open Access
Phenotypic reversion and target prioritization for cellular inflammation via representation learning with foundation models
The identification of genetic perturbations that can reverse disease-associated cellular phenotypes toward a healthy state is a central challenge in early drug discovery. We present a proof-of-concept framework leveraging single-cell foundation models (scFMs) and a large-scale Perturb-seq dataset to prioritize targets for phenotypic reversion of cellular inflammation. We incorporated both basal and proinflammatory signaling conditions, specifically stimulation with interleukin-1 beta (IL-1β) and tumor necrosis factor alpha (TNF-α), to assess whether atherosclerotic disease-relevant stimulation improved identification of genes and pathways critical to disease progression. Our dataset comprised 864,115 endothelial cells subjected to 1,740 unique genetic perturbations. Through having both conditions, we identified targets that exhibited differential effects on gene expression dependent on cellular state. Using scFMs, we embedded single-cell transcriptomes into high-dimensional latent spaces and ranked perturbations by their ability to shift the inflammatory transcriptomic profile toward that of untreated controls. Benchmarking against both annotated gene sets and expert-curated targets, we found robust enrichment for known regulators of inflammation and biologically relevant targets, despite these models having no prior information about these targets. Importantly, including both basal and proinflammatory conditions improved identification of inflammation-associated targets compared to using just the basal condition. This underscored the value of incorporating disease-relevant stimulations in perturbation experiments. Our results highlight the utility of scFMs for data-driven target nomination, emphasize the role of cellular state in regulatory responses, and provide a scalable, model-agnostic approach for ML-guided target discovery. This work offers a valuable community resource for advancing biological understanding of inflammatory-associated disease.
Open AccessPhenotypic reversion and target prioritization for cellular inflammation via representation learning with foundation models
The identification of genetic perturbations that can reverse disease-associated cellular phenotypes toward a healthy state is a central challenge in early drug discovery. We present a proof-of-concept framework leveraging single-cell foundation models (scFMs) and a large-scale Perturb-seq dataset to prioritize targets for phenotypic reversion of cellular inflammation. We incorporated both basal and proinflammatory signaling conditions, specifically stimulation with interleukin-1 beta (IL-1β) and tumor necrosis factor alpha (TNF-α), to assess whether atherosclerotic disease-relevant stimulation improved identification of genes and pathways critical to disease progression. Our dataset comprised 864,115 endothelial cells subjected to 1,740 unique genetic perturbations. Through having both conditions, we identified targets that exhibited differential effects on gene expression dependent on cellular state. Using scFMs, we embedded single-cell transcriptomes into high-dimensional latent spaces and ranked perturbations by their ability to shift the inflammatory transcriptomic profile toward that of untreated controls. Benchmarking against both annotated gene sets and expert-curated targets, we found robust enrichment for known regulators of inflammation and biologically relevant targets, despite these models having no prior information about these targets. Importantly, including both basal and proinflammatory conditions improved identification of inflammation-associated targets compared to using just the basal condition. This underscored the value of incorporating disease-relevant stimulations in perturbation experiments. Our results highlight the utility of scFMs for data-driven target nomination, emphasize the role of cellular state in regulatory responses, and provide a scalable, model-agnostic approach for ML-guided target discovery. This work offers a valuable community resource for advancing biological understanding of inflammatory-associated disease.
- Open Access
- 23 May 2025
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 41  (6)
Simple controls exceed best deep learning algorithms and reveal foundation model effectiveness for predicting genetic perturbations
Motivation: Modeling genetic perturbations and their effect on the transcriptome is a key area of pharmaceutical research. Due to the complexity of the transcriptome, there has been much excitement and development in deep learning (DL) because of its ability to model complex relationships. In particular, the transformer-based foundation model paradigm emerged as the gold-standard of predicting post-perturbation responses. However, understanding these increasingly complex models and evaluating their practical utility is lacking, along with simple but appropriate benchmarks to compare predictive methods. Results: Here, we present a simple baseline method that outperforms both state of the art (SOTA) in DL and other proposed simpler neural architectures, setting a necessary benchmark to evaluate in the field of post-perturbation prediction. We also elucidate the utility of foundation models for the task of post-perturbation prediction via generalizable fine-tuning experiments that can be translated to different applications of transformer-based foundation models to tasks of interest. Furthermore, we provide a corrected version of a popular dataset used for benchmarking perturbation prediction models. Our hope is that this work will properly contextualize further development of DL models in the perturbation space with necessary control procedures. Availability and implementation: All source code is available at: https://github.com/pfizer-opensource/perturb_seq. The DOI is 10.5281/zenodo.15352937.
- 23 May 2025
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 41  (6)
Open AccessSimple controls exceed best deep learning algorithms and reveal foundation model effectiveness for predicting genetic perturbations
Motivation: Modeling genetic perturbations and their effect on the transcriptome is a key area of pharmaceutical research. Due to the complexity of the transcriptome, there has been much excitement and development in deep learning (DL) because of its ability to model complex relationships. In particular, the transformer-based foundation model paradigm emerged as the gold-standard of predicting post-perturbation responses. However, understanding these increasingly complex models and evaluating their practical utility is lacking, along with simple but appropriate benchmarks to compare predictive methods. Results: Here, we present a simple baseline method that outperforms both state of the art (SOTA) in DL and other proposed simpler neural architectures, setting a necessary benchmark to evaluate in the field of post-perturbation prediction. We also elucidate the utility of foundation models for the task of post-perturbation prediction via generalizable fine-tuning experiments that can be translated to different applications of transformer-based foundation models to tasks of interest. Furthermore, we provide a corrected version of a popular dataset used for benchmarking perturbation prediction models. Our hope is that this work will properly contextualize further development of DL models in the perturbation space with necessary control procedures. Availability and implementation: All source code is available at: https://github.com/pfizer-opensource/perturb_seq. The DOI is 10.5281/zenodo.15352937.
- Open Access
Simple controls exceed best deep learning algorithms and reveal foundation model effectiveness for predicting genetic perturbations
Modeling genetic perturbations and their effect on the transcriptome is a key area of pharmaceutical research. Due to the complexity of the transcriptome, there has been much excitement and development in deep learning (DL) because of its ability to model complex relationships. In particular, the transformer-based foundation model paradigm emerged as the gold-standard of predicting post-perturbation responses. However, understanding these increasingly complex models and evaluating their practical utility is lacking, along with simple but appropriate benchmarks to compare predictive methods. Here, we present a simple baseline method that outperforms both state of the art (SOTA) in DL and other proposed simpler neural architectures, setting a necessary benchmark to evaluate in the field of post-perturbation prediction. We also elucidate the utility of foundation models for the task of post-perturbation prediction via generalizable fine-tuning experiments that can be translated to different applications of transformer-based foundation models to tasks of interest. Furthermore, we provide a corrected version of a popular dataset used for benchmarking perturbation prediction models. Our hope is that this work will properly contextualize further development of DL models in the perturbation space with necessary control procedures.
Open AccessSimple controls exceed best deep learning algorithms and reveal foundation model effectiveness for predicting genetic perturbations
Modeling genetic perturbations and their effect on the transcriptome is a key area of pharmaceutical research. Due to the complexity of the transcriptome, there has been much excitement and development in deep learning (DL) because of its ability to model complex relationships. In particular, the transformer-based foundation model paradigm emerged as the gold-standard of predicting post-perturbation responses. However, understanding these increasingly complex models and evaluating their practical utility is lacking, along with simple but appropriate benchmarks to compare predictive methods. Here, we present a simple baseline method that outperforms both state of the art (SOTA) in DL and other proposed simpler neural architectures, setting a necessary benchmark to evaluate in the field of post-perturbation prediction. We also elucidate the utility of foundation models for the task of post-perturbation prediction via generalizable fine-tuning experiments that can be translated to different applications of transformer-based foundation models to tasks of interest. Furthermore, we provide a corrected version of a popular dataset used for benchmarking perturbation prediction models. Our hope is that this work will properly contextualize further development of DL models in the perturbation space with necessary control procedures.
- Open Access
- 16 August 2023
- Published by Royal Society of Chemistry (RSC) in Digital Discovery
- Vol. 2  (5) , 1354-1367
Deep representation learning determines drug mechanism of action from cell painting images
Fluorescent-based microscopy screens carry a broad range of phenotypic information about how compounds affect cellular biology. From changes in cellular morphology observed in these screens, one key area of medicinal interest is determining a compound's mechanism of action. However, much of this phenotypic information is subtle and difficult to quantify. Hence, creating quantitative embeddings that can measure cellular response to compound perturbation has been a key area of research. Here we present a deep learning enabled encoder called MOAProfiler that captures phenotypic features for determining mechanism of action from Cell Painting images. We compared our method with both the traditional gold-standard means of feature encoding via CellProfiler and a deep learning encoder called DeepProfiler. The results, on two independent and biologically different datasets, indicated that MOAProfiler encoded MOA-specific features that allowed for more accurate clustering and classification of compounds over hundreds of different MOAs.
- 16 August 2023
- Published by Royal Society of Chemistry (RSC) in Digital Discovery
- Vol. 2  (5) , 1354-1367
Open AccessDeep representation learning determines drug mechanism of action from cell painting images
Fluorescent-based microscopy screens carry a broad range of phenotypic information about how compounds affect cellular biology. From changes in cellular morphology observed in these screens, one key area of medicinal interest is determining a compound's mechanism of action. However, much of this phenotypic information is subtle and difficult to quantify. Hence, creating quantitative embeddings that can measure cellular response to compound perturbation has been a key area of research. Here we present a deep learning enabled encoder called MOAProfiler that captures phenotypic features for determining mechanism of action from Cell Painting images. We compared our method with both the traditional gold-standard means of feature encoding via CellProfiler and a deep learning encoder called DeepProfiler. The results, on two independent and biologically different datasets, indicated that MOAProfiler encoded MOA-specific features that allowed for more accurate clustering and classification of compounds over hundreds of different MOAs.
- Open Access
- 24 June 2023
- Published by Springer Nature in Communications Biology
- Vol. 6  (1) , 1-9
Learning fast and fine-grained detection of amyloid neuropathologies from coarse-grained expert labels
Precise, scalable, and quantitative evaluation of whole slide images is crucial in neuropathology. We release a deep learning model for rapid object detection and precise information on the identification, locality, and counts of cored plaques and cerebral amyloid angiopathy (CAA). We trained this object detector using a repurposed image-tile dataset without any human-drawn bounding boxes. We evaluated the detector on a new manually-annotated dataset of whole slide images (WSIs) from three institutions, four staining procedures, and four human experts. The detector matched the cohort of neuropathology experts, achieving 0.64 (model) vs. 0.64 (cohort) average precision (AP) for cored plaques and 0.75 vs. 0.51 AP for CAAs at a 0.5 IOU threshold. It provided count and locality predictions that approximately correlated with gold-standard human CERAD-like WSI scoring (p = 0.07 ± 0.10). The openly-available model can quickly score WSIs in minutes without a GPU on a standard workstation.
- 24 June 2023
- Published by Springer Nature in Communications Biology
- Vol. 6  (1) , 1-9
Open AccessLearning fast and fine-grained detection of amyloid neuropathologies from coarse-grained expert labels
Precise, scalable, and quantitative evaluation of whole slide images is crucial in neuropathology. We release a deep learning model for rapid object detection and precise information on the identification, locality, and counts of cored plaques and cerebral amyloid angiopathy (CAA). We trained this object detector using a repurposed image-tile dataset without any human-drawn bounding boxes. We evaluated the detector on a new manually-annotated dataset of whole slide images (WSIs) from three institutions, four staining procedures, and four human experts. The detector matched the cohort of neuropathology experts, achieving 0.64 (model) vs. 0.64 (cohort) average precision (AP) for cored plaques and 0.75 vs. 0.51 AP for CAAs at a 0.5 IOU threshold. It provided count and locality predictions that approximately correlated with gold-standard human CERAD-like WSI scoring (p = 0.07 ± 0.10). The openly-available model can quickly score WSIs in minutes without a GPU on a standard workstation.
- Open Access
Learning fast and fine-grained detection of amyloid neuropathologies from coarse-grained expert labels
Precise, scalable, and quantitative evaluation of whole slide images is crucial in neuropathology. We release a deep learning model for rapid object detection and precise information on the identification, locality, and counts of cored plaques and cerebral amyloid angiopathies (CAAs). We trained this object detector using a repurposed image-tile dataset without any human-drawn bounding boxes. We evaluated the detector on a new manually-annotated dataset of whole slide images (WSIs) from three institutions, four staining procedures, and four human experts. The detector matched the cohort of neuropathology experts, achieving 0.64 (model) vs. 0.64 (cohort) average precision (AP) for cored plaques and 0.75 vs. 0.51 AP for CAAs at a 0.5 IOU threshold. It provided count and locality predictions that correlated with gold-standard CERAD-like WSI scoring (p=0.07± 0.10). The openly-available model can quickly score WSIs in minutes without a GPU on a standard workstation.
Open AccessLearning fast and fine-grained detection of amyloid neuropathologies from coarse-grained expert labels
Precise, scalable, and quantitative evaluation of whole slide images is crucial in neuropathology. We release a deep learning model for rapid object detection and precise information on the identification, locality, and counts of cored plaques and cerebral amyloid angiopathies (CAAs). We trained this object detector using a repurposed image-tile dataset without any human-drawn bounding boxes. We evaluated the detector on a new manually-annotated dataset of whole slide images (WSIs) from three institutions, four staining procedures, and four human experts. The detector matched the cohort of neuropathology experts, achieving 0.64 (model) vs. 0.64 (cohort) average precision (AP) for cored plaques and 0.75 vs. 0.51 AP for CAAs at a 0.5 IOU threshold. It provided count and locality predictions that correlated with gold-standard CERAD-like WSI scoring (p=0.07± 0.10). The openly-available model can quickly score WSIs in minutes without a GPU on a standard workstation.
- Open Access
Deep Representation Learning Determines Drug Mechanism of Action from Cell Painting Images
Fluorescent-based microscopy screens carry a broad range of phenotypic information about how compounds affect cellular biology. From changes in cellular morphology observed in these screens, one key area of medicinal interest is determining a compound’s mechanism of action. However, much of this phenotypic information is subtle and difficult to quantify. Hence, creating quantitative embeddings that can measure cellular response to compound perturbation has been a key area of research. Here we present a deep learning enabled encoder called MOAProfiler that captures phenotypic features for determining mechanism of action from Cell Painting images. We compared our method with both a traditional computer vision means of feature encoding via CellProfiler and a deep learning encoder called DeepProfiler. The results, on two independent and biologically different datasets, indicated that MOAProfiler encoded MOA-specific features that allowed for more accurate clustering and classification of compounds over hundreds of different MOAs.
Open AccessDeep Representation Learning Determines Drug Mechanism of Action from Cell Painting Images
Fluorescent-based microscopy screens carry a broad range of phenotypic information about how compounds affect cellular biology. From changes in cellular morphology observed in these screens, one key area of medicinal interest is determining a compound’s mechanism of action. However, much of this phenotypic information is subtle and difficult to quantify. Hence, creating quantitative embeddings that can measure cellular response to compound perturbation has been a key area of research. Here we present a deep learning enabled encoder called MOAProfiler that captures phenotypic features for determining mechanism of action from Cell Painting images. We compared our method with both a traditional computer vision means of feature encoding via CellProfiler and a deep learning encoder called DeepProfiler. The results, on two independent and biologically different datasets, indicated that MOAProfiler encoded MOA-specific features that allowed for more accurate clustering and classification of compounds over hundreds of different MOAs.
- Open Access
- 28 April 2022
- Published by Springer Nature in Acta Neuropathologica Communications
- Vol. 10  (1) , 1-22
Deep learning from multiple experts improves identification of amyloid neuropathologies
Pathologists can label pathologies differently, making it challenging to yield consistent assessments in the absence of one ground truth. To address this problem, we present a deep learning (DL) approach that draws on a cohort of experts, weighs each contribution, and is robust to noisy labels. We collected 100,495 annotations on 20,099 candidate amyloid beta neuropathologies (cerebral amyloid angiopathy (CAA), and cored and diffuse plaques) from three institutions, independently annotated by five experts. DL methods trained on a consensus-of-two strategy yielded 12.6–26% improvements by area under the precision recall curve (AUPRC) when compared to those that learned individualized annotations. This strategy surpassed individual-expert models, even when unfairly assessed on benchmarks favoring them. Moreover, ensembling over individual models was robust to hidden random annotators. In blind prospective tests of 52,555 subsequent expert-annotated images, the models labeled pathologies like their human counterparts (consensus model AUPRC = 0.74 cored; 0.69 CAA). This study demonstrates a means to combine multiple ground truths into a common-ground DL model that yields consistent diagnoses informed by multiple and potentially variable expert opinions.
- 28 April 2022
- Published by Springer Nature in Acta Neuropathologica Communications
- Vol. 10  (1) , 1-22
Open AccessDeep learning from multiple experts improves identification of amyloid neuropathologies
Pathologists can label pathologies differently, making it challenging to yield consistent assessments in the absence of one ground truth. To address this problem, we present a deep learning (DL) approach that draws on a cohort of experts, weighs each contribution, and is robust to noisy labels. We collected 100,495 annotations on 20,099 candidate amyloid beta neuropathologies (cerebral amyloid angiopathy (CAA), and cored and diffuse plaques) from three institutions, independently annotated by five experts. DL methods trained on a consensus-of-two strategy yielded 12.6–26% improvements by area under the precision recall curve (AUPRC) when compared to those that learned individualized annotations. This strategy surpassed individual-expert models, even when unfairly assessed on benchmarks favoring them. Moreover, ensembling over individual models was robust to hidden random annotators. In blind prospective tests of 52,555 subsequent expert-annotated images, the models labeled pathologies like their human counterparts (consensus model AUPRC = 0.74 cored; 0.69 CAA). This study demonstrates a means to combine multiple ground truths into a common-ground DL model that yields consistent diagnoses informed by multiple and potentially variable expert opinions.
- Open Access
Deep learning from multiple experts improves identification of amyloid neuropathologies
Pathologists can label pathologies differently, making it challenging to yield consistent assessments in the absence of one ground truth. To address this problem, we present a DL approach that draws on a cohort of experts, weighs each contribution, and is robust to noisy labels. We collected 100,495 annotations on 20,099 candidate amyloid beta neuropathologies (cerebral amyloid angiopathy (CAA), and cored and diffuse plaques) from three institutions, independently annotated by five experts. DL methods trained on a consensus-of-two strategy yielded 12.6-26% improvements by area under the precision recall curve (AUPRC) when compared to those that learned individualized annotations. This strategy surpassed individual-expert models, even when unfairly assessed on benchmarks favoring them. Moreover, ensembling over individual models was robust to hidden random annotators. In blind prospective tests of 52,555 subsequent expert-annotated images, the models labeled pathologies like their human counterparts (consensus model AUPRC=0.74 cored; 0.69 CAA). This study demonstrates a means to combine multiple ground truths into a common-ground DL model that yields consistent diagnoses informed by multiple and potentially variable expert opinion.
Open AccessDeep learning from multiple experts improves identification of amyloid neuropathologies
Pathologists can label pathologies differently, making it challenging to yield consistent assessments in the absence of one ground truth. To address this problem, we present a DL approach that draws on a cohort of experts, weighs each contribution, and is robust to noisy labels. We collected 100,495 annotations on 20,099 candidate amyloid beta neuropathologies (cerebral amyloid angiopathy (CAA), and cored and diffuse plaques) from three institutions, independently annotated by five experts. DL methods trained on a consensus-of-two strategy yielded 12.6-26% improvements by area under the precision recall curve (AUPRC) when compared to those that learned individualized annotations. This strategy surpassed individual-expert models, even when unfairly assessed on benchmarks favoring them. Moreover, ensembling over individual models was robust to hidden random annotators. In blind prospective tests of 52,555 subsequent expert-annotated images, the models labeled pathologies like their human counterparts (consensus model AUPRC=0.74 cored; 0.69 CAA). This study demonstrates a means to combine multiple ground truths into a common-ground DL model that yields consistent diagnoses informed by multiple and potentially variable expert opinion.