April ABC Seminar: Maria Carilli – Caltech – “Biophysical modeling with variational autoencoders for bimodal, single-cell RNA sequencing data”

April ABC Seminar: Maria Carilli – Caltech – “Biophysical modeling with variational autoencoders for bimodal, single-cell RNA sequencing data”

Transcription in individual cells is an inherently stochastic process, with non-deterministic numbers of RNA molecules produced per gene in transcriptional events across cells of the same cell type and condition. These distributions can be mathematical modeled according to biophysical hypotheses that specify system parameters of interest, such as rates of RNA splicing and degradation. While such stochastic models have been applied to infer RNA processing rates in fluorescence based assays, the scale of high-throughput sequencing data makes inference in this setting computationally challenging, especially as more molecular species, such as unspliced and spliced RNA molecules, can be quantified. We adapt an inference tool from machine learning, the variational autoencoder, that allows summarization and analysis of high-dimensional datasets to work with biophysical models of transcriptional dynamics. While preserving the variational autoencoder’s ability to reduce dimensionality of and denoise data, this work presents two major advances: one, principled integration and treatment of multimodal data (i.e., unspliced and spliced RNA molecules) through the employment of specific biophysical models; and two, inference of biophysical rates that parameterize those models, allowing the exploration of RNA dynamics across thousands of individual genes and cells.

Speaker:  Maria Carilli
Affiliation: California Institute of Technology
Position:  PhD Student, Lior Pachter Lab
Host: Dan MacDonald, PhD, Gibson Lab

Date: Monday April 28th, 2025
Time: 4:00-5:00PM ET
Zoom: https://partners.zoom.us/j/82163676866
Meeting ID: 821 6367 6866

Relevant Papers: Biophysical modeling with variational autoencoders for bimodal, single-cell RNA sequencing data, Spectral neural approximations for models of transcriptional dynamics

 Maria Carilli is a graduate student at Caltech working on theory and software development for the analysis of single-cell RNA sequencing data. She obtained undergraduate degrees in biophysics and music from the University of Colorado Boulder with a minor in computational biology. She is particularly interested in combining tools from machine learning with mathematical modeling of biophysical dynamics for deeper, and scalable, understanding of cellular systems.

Click here to be added to our mail list.

 

 

 

February ABC Seminar: Prateek Prasanna, PhD – Stony Brook University – “Reality-aware Medical Vision: The Quest for Interpretable and Explainable AI in Medical Imaging”

February ABC Seminar: Prateek Prasanna, PhD – Stony Brook University – “Reality-aware Medical Vision: The Quest for Interpretable and Explainable AI in Medical Imaging”

Prateek Prasanna, PhDMedical imaging plays a critical role in modern healthcare, yet its complexity necessitates advanced computational tools to enhance analysis, interpretability, and diagnostic accuracy. In this presentation, we will discuss our research efforts in developing computational imaging biomarkers and frameworks for precision medicine in real clinical scenarios involving imperfect data. We will cover a spectrum of computational techniques grounded in both biological and domain-specific insights that facilitate the early detection and evaluation of treatment responses across various diseases.

In radiology, we have developed methods to analyze intricate tissue structures using radiomics and topology-informed deep learning, demonstrating improved prediction of treatment response in multiple cancers. Additionally, we explore techniques that integrate expert eye gaze patterns and radiomic features to condition diffusion models, enabling disease-aware image synthesis and enhancing the generation of anatomically accurate and clinically relevant medical images. These approaches provide both diagnostic accuracy and interpretability, bridging the gap between AI models and clinical needs.

In digital pathology, our work focuses on overcoming challenges such as domain shifts, data scarcity, and limited interpretability. We employ adaptive meta-learning frameworks to generalize across unseen staining patterns, improving segmentation and classification of histopathology images. Generative models are utilized to synthesize high-resolution pathology images and restore critical details, leveraging pathology text reports and handcrafted features to enhance both quality and interpretability. Finally, we integrate expert pathological insights with deep learning to improve interpretability in whole-slide image analysis, demonstrating the potential to provide meaningful explanations alongside robust performance.

These advancements highlight the transformative potential of explainable and generative AI in radiology and digital pathology, paving the way for innovative, clinically relevant solutions in medical imaging.

Speaker:  Prateek Prasanna, PhD
Affiliation: Stony Brook University
Position:  Assistant Professor, Department of Biomedical Informatics
Host: Daniel Shao, Mahmood Lab

Date: Monday February 24th, 2025
Time: 4:00-5:00PM ET
Zoom: https://partners.zoom.us/j/82163676866
Meeting ID: 821 6367 6866

Research Links: Publications, IMAGINE Lab

Prateek Prasanna is an Assistant Professor in the Department of Biomedical Informatics at Stony Brook University and directs the Imaging Informatics for Precision Medicine (IMAGINE) Lab. He received his PhD in Biomedical Engineering from Case Western Reserve University, Ohio, USA. Prior to that, he obtained his master’s degree in Electrical and Computer Engineering from Rutgers University and bachelor’s degree in Electrical and Electronics Engineering from National Institute of Technology, Calicut, India. Dr. Prasanna’s research focuses on building clinically translatable machine learning tools that leverage multiple data streams of imaging, pathology, and genomics to derive actionable insights for enabling better treatment decisions. His research involving development of companion diagnostic tools for thoracic, neuro, and breast imaging applications has been published in venues such as MICCAI, CVPR, ECCV, NeurIPS, Radiology, Medical Image Analysis, etc, and has won several innovation awards. One of the core focuses of his lab is to integrate machine generated inferences with expert clinical reads to make clinical workflows more efficient and effective. His team has been actively working on the advancement of interpretable machine learning and xAI techniques to facilitate the discovery of computational biomarkers, particularly in situations where data is limited or missing.

Click here to be added to our mail list.

January ABC Seminar: Stephen Pfohl, PhD – Google Research – “Algorithmic fairness and responsible AI for health equity”

January ABC Seminar: Stephen Pfohl, PhD – Google Research – “Algorithmic fairness and responsible AI for health equity”

Stephen PfohlIn this talk, I aim to present insights into the design of evaluations of machine learning and AI systems to assess properties related to algorithmic fairness and health equity. I argue that such evaluations are contextual and require specification of the intended use, target population, and assumptions regarding the data generating process and measurement mechanism. Through examples from my research, I argue that causal graphical models can serve as key tools for context specification and can aid in understanding and appropriate use of analytical algorithmic fairness techniques. I will further discuss recent work related to evaluation of equity-related biases in large language models.

Speaker:  Stephen Pfohl, PhD
Affiliation: Google Research
Position:  Senior Research Scientist
Host: Anurag Vaidya, Mahmood Lab

Date: Monday January 27, 2025
Time: 4:00-5:00PM ET
Zoom: https://partners.zoom.us/j/82163676866
Meeting ID: 821 6367 6866

Research Links: Personal siteGoogle Scholar

Stephen Pfohl is a senior research scientist at Google Research. His work focuses on the incorporation of fairness, distribution shift, and equity considerations into the design and evaluation of machine learning systems in healthcare contexts. Stephen earned his PhD in Biomedical Informatics from Stanford University.

November ABC Seminar: Brian Hie, PhD – Stanford University – “Sequence modeling from molecular to genome scale with Evo”

November ABC Seminar: Brian Hie, PhD – Stanford University – “Sequence modeling from molecular to genome scale with Evo”

The genome is a sequence that encodes the DNA, RNA, and proteins orchestrating an organism’s function. We present Evo, a long-context genomic foundation model with a frontier architecture trained on millions of prokaryotic and phage genomes, and report the first scaling laws on DNA to complement observations in language and vision. Evo generalizes across DNA, RNA, and proteins, enabling zero-shot function prediction competitive with domain-specific language models and the generation of functional CRISPR-Cas and transposon systems, representing the first examples of protein-RNA and protein-DNA co-design with a language model. Evo also learns how small mutations affect whole-organism fitness and generates megabase-scale sequences with plausible genomic architecture. These prediction and generation capabilities span molecular to genome scales of complexity, advancing our understanding and control of biology.

Title: Sequence modeling from molecular to genome scale with Evo
Speaker:  Brian Hie, PhD
Affiliation: Stanford University, Chemical Engineering and Data Science
Position:  Assistant Professor
Host: Jiening Zhu, PhD – Gerber Lab

Date: Monday November 18, 2024
Time: 4:00-5:00PM ET
Zoom: https://partners.zoom.us/j/82163676866
Meeting ID: 821 6367 6866

Brian Hie is an Assistant Professor of Chemical Engineering at Stanford University, the Dieter Schwarz Foundation Stanford Data Science Faculty Fellow, and an Innovation Investigator at Arc Institute, where his group conducts research at the intersection of biology and machine learning.

Click here to be added to our mail list.

October ABC Seminar: Zhi Huang, PhD – Univ of Pennsylvania – “A pathologist–AI collaboration framework for enhancing diagnostic accuracies and efficiencies”

October ABC Seminar: Zhi Huang, PhD – Univ of Pennsylvania – “A pathologist–AI collaboration framework for enhancing diagnostic accuracies and efficiencies”

The integration of Artificial Intelligence (AI) in clinical pathology has faced significant hurdles due to constraints in data collection and challenges associated with model transparency and interpretability.  In this talk, we introduce a novel digital pathology AI framework named nuclei.io, which leverages active learning and incorporates real-time feedback from human experts. This innovative approach empowers pathologists to quickly generate diverse datasets and develop models for various clinical applications. To demonstrate the effectiveness of our framework, we conducted two user studies employing such human–AI collaboration strategy. These studies focused on two key areas: the identification of plasma cells in endometrial biopsies and the detection of colorectal cancer metastasis in lymph nodes. The results from these studies showed significant enhancements in sensitivity, accuracy, and diagnostic efficiency with the integration of AI. Our findings underscore the benefits of the human-in-the-loop AI framework, highlighting its potential to transform the field of digital pathology.

Speaker:  Zhi Huang, PhD
Affiliation: Perelman School of Medicine, University of Pennsylvania
Position:  Assistant Professor (incoming), Dept of Pathology and Laboratory Medicine, Dept of Biostatistics, Epidemiology and Informatics
Research Links: https://www.zhihuang.ai
Host: Andrew Song, PhD – Mahmood Lab

Date: Monday, October 21, 2024
Time: 4:00-5:00PM ET
Zoom: https://partners.zoom.us/j/82163676866
Meeting ID: 821 6367 6866

Zhi Huang is an incoming tenure-track assistant professor at the University of Pennsylvania starting January 2025. He obtained his PhD in Electrical and Computer Engineering (ECE) from Purdue University in August 2021. Since August 2021, He has been a postdoctoral fellow at Stanford University. His research focuses on AI/ML innovation and its application to medicine, with topics including vision-language foundation model for pathology, human-AI collaboration, neurodegenerative diseases, etc. His research has drawn wide public attention (including the New York Times, Stanford Magazine, and Stanford Scope) and has resulted in translational innovations. In 2022, Zhi Huang co-founded nuclei.io — a human-in-the-loop AI platform for digital pathology. It was selected as one of only 9 Stanford Catalyst 2023 cohort innovations.

Click here to be added to our mail list.

ABC Hybrid Seminar: Bokai Zhu, PhD – Ragon Institute – “Integration of spatial-omics and single-cell data across modalities with weakly linked features”

ABC Hybrid Seminar: Bokai Zhu, PhD – Ragon Institute – “Integration of spatial-omics and single-cell data across modalities with weakly linked features”

Bokai ZhuAdvancements in single-cell and spatial-omics technologies have created a need for integrating datasets across modalities with limited and weakly correlated features, such as those between spatial proteomics and transcriptomics. Existing tools, usually designed for strongly linked data, often fail in these scenarios. Recently, we have developed a series of methods (Mario and MaxFuse), that improves integration by refining weak correlations between modalities through an iterative smoothing and co-embedding process, and achieves single-cell level matching across these weakly linked modalities, enabling in-depth understanding of tissue micro-environments.

Speaker:  Bokai Zhu, PhD
Affiliation: Ragon Institute of MGH, Harvard and MIT, Broad Institute, MIT
Position:  Postdoctoral Research Fellow
Research Links: https://bokaizhu.github.io/
Host: Muhammad Shaban, PhD – Mahmood Lab

Date: Monday, September 23, 2024
Time: 1:00-2:00PM ET
In-person: Duncan Reid Conference Room (directions below)
Zoom: https://partners.zoom.us/j/82163676866
Meeting ID: 821 6367 6866

Bokai Zhu is currently a postdoctoral researcher supervised by Prof. Alex Shalek at the Ragon Institute at MGH, MIT, and Harvard. Prior to that, Dr. Zhu obtained his PhD in Microbiology and Immunology from Stanford University, under the supervision of Prof. Garry Nolan. He received a bachelor’s degree in Biology from Cornell University and Zhejiang University. Dr. Zhu’s doctoral research focused on: 1) Experimental assay development for multiplex imaging platforms; 2) Computational algorithm development for single-cell multi-omic integrations; 3)Application of the above tools in various biological systems.

Click here to be added to our mail list.

May ABC Seminar: Jean du Terrail, PhD – Owkin – “Federated Learning in Healthcare in the Real-World: Examples and Practical Challenges”

May ABC Seminar: Jean du Terrail, PhD – Owkin – “Federated Learning in Healthcare in the Real-World: Examples and Practical Challenges”

With the never-ending revolutions of data-driven approaches that started already more than a decade ago, it is surprising at first that the pace of ML discoveries in medicine seems substantially slower than the ones found in consumer applications such as ChatGPT. The answer to that paradox is relatively simple: in healthcare, data is hard to access as it is expensive and siloed in medical institutions and thus data-hungry methods trained on only one center on limited data often fail to generalize to another. In order to break those silos while protecting patient data and thus enabling new medical discoveries via ML, federated learning (FL) is a promising approach. However, in practice, applying FL in real-life contexts presents numerous challenges. This talk will use some of the FL research projects that Owkin has spearheaded in order to illustrate the realities of FL in healthcare and discuss the remaining milestones on the road to FL technologies and ML becoming the new engine of medicine research.

Special Seminar: Anshul Kundaje, PhD – “Using deep learning models to debug regulatory genomics experiments and decode cis-regulatory syntax”

Special Seminar: Anshul Kundaje, PhD – “Using deep learning models to debug regulatory genomics experiments and decode cis-regulatory syntax”

BWH Computational Pathology Special Seminar

Title: Using deep learning models to debug regulatory genomics experiments and decode cis-regulatory syntax
Speaker: Anshul Kundaje, PhD
Affiliation: Stanford University
Position: Associate Professor, Genetics and Computer Science

Date: Monday April 22, 2024
Time: 4:00PM-5:00PM ET
Zoom: https://partners.zoom.us/j/82163676866
Meeting ID: 821 6367 6866

Anshul Kundaje, PhD, is Associate Professor of Genetics and Computer Science at Stanford University. His primary research area is large-scale computational regulatory genomics. The Kundaje lab develops deep learning models of gene regulation and model interpretation methods to decipher non-coding DNA and genetic variation associated with disease. Dr. Kundaje has led computational efforts to develop widely used resources in collaboration with several NIH consortia including ENCODE, Roadmap Epigenomics and IGVF. Dr. Kundaje is a recipient of the 2016 NIH Director’s New Innovator Award and the 2014 Alfred Sloan Fellowship.

Links: The Encyclopedia of DNA Elements (ENCODE) Project, Stanford University, MIT

March ABC Seminar: Andrew H. Song – Brigham and Women’s Hospital – “Towards 3D pathology – The opportunities and challenges”

March ABC Seminar: Andrew H. Song – Brigham and Women’s Hospital – “Towards 3D pathology – The opportunities and challenges”

Human tissue, which is inherently three-dimensional (3D), is traditionally examined through standard-of-care histopathology as limited two-dimensional (2D) cross sections that can poorly represent the tissue due to sampling bias. To holistically characterize 3D histomorphology, 3D imaging modalities have been developed, but clinical translation is hampered by the complex and time-consuming requirements for manual evaluation, as well as the current lack of computational platforms to distill clinical insights from these large, high-resolution datasets. We present a deep learning model for processing tissue volumes and predicting patient outcomes with weak supervision. Recurrence risk-stratification models were trained with archived prostate cancer specimens imaged with open-top light-sheet microscopy or microcomputed tomography. By comprehensively capturing 3D morphologies, 3D block-based prognostication achieves superior performance to traditional 2D slice-based approaches, including existing clinical/histopathological baselines. Incorporating larger tissue volumes is shown to improve prognostic accuracy. This framework offers a promising direction for clinical decision support and 3D biomarker discovery, with the potential to further catalyze the growth of 3D spatial biology techniques for clinical applications.

Publication: Song AH et al., “Weakly supervised AI for efficient analysis of 3D pathology samples”, Cell (2024, In Press) Preprint

Speaker: Andrew H. Song, PhD
Affiliation: Brigham and Women’s Hospital, Harvard Medical School
Position: Research Fellow

Date: Monday, March 25, 2024
Time: 4:00PM-5:00PM ET

Zoom: https://partners.zoom.us/j/82163676866
Meeting ID: 821 6367 6866

In Person: VTC-2006b, Hale BTM 2nd Floor
Brigham and Women’s Hospital
60 Fenwood Rd, Boston 02115

Andrew is currently a postdoctoral research fellow at Brigham and Women’s Hospital and Harvard Medical School, working with Prof. Faisal Mahmood in the AI4Pathology group since early 2022. His current research focus is developing DL-based frameworks for 3D computational pathology, using images from diverse 3D imaging modalities. In his previous life, he received a Ph.D. from Massachusetts Institute of Technology (MIT) Electrical Engineering and Computer Science (EECS), co-advised by Prof. Emery N. Brown (MIT) and Prof. Demba Ba (Harvard), working at the intersection of computational neuroscience and statistical signal processing.

February ABC Seminar: Vitalii Kleshchevnikov, PhD – Wellcome Sanger Institute – “Probabilistic models to resolve cell identity and tissue architecture”

February ABC Seminar: Vitalii Kleshchevnikov, PhD – Wellcome Sanger Institute – “Probabilistic models to resolve cell identity and tissue architecture”

Cell identity drives cell-cell communication and tissue architecture and is in return regulated by cell-extrinsic cues. Cell identity is determined by the combination of intrinsic developmentally established transcription factor use (TF) and constitutive as well as cell communication-dependent TF activities. Presented work shows two probabilistic models that we developed to advance the understanding of these processes using single-cell and spatial genomic data.

Spatial transcriptomic technologies promise to resolve cellular wiring diagrams of tissues in health and disease, but comprehensive mapping of cell types in situ remains a challenge. Here we present cell2location, a Bayesian model that can resolve fine-grained cell types in spatial transcriptomic data and create comprehensive cellular maps of diverse tissues. Cell2location accounts for technical sources of variation and borrows statistical strength across locations, thereby enabling the integration of single cell and spatial transcriptomics with higher sensitivity and resolution than existing tools. We assess cell2location in three different tissues and demonstrate improved mapping of fine-grained cell types. In the mouse brain, we discover fine regional astrocyte subtypes across the thalamus and hypothalamus. In the human lymph node, we spatially map a rare pre-germinal center B cell population. In the human gut, we resolve fine immune cell populations in lymphoid follicles. Collectively our results present cell2location as a versatile analysis tool for mapping tissue architectures in a comprehensive manner.

Python package is provided here:  https://github.com/BayraktarLab/cell2location.

Cell identity and plasticity is regulated by a combinatorial code mediated by transcription factors and the cell communication environment. Systematically dissecting how the regulatory code robustly defines the vast complexity of cell populations across tissues is a long-standing challenge. Measured using the assay for transposase-accessible chromatin with sequencing (ATAC-seq), DNA accessibility provides a readout of intermediate gene regulation steps at single-cell resolution, with technologies measuring both RNA and ATAC providing the necessary evidence to build mechanistic models of regulation. Existing methods address one or several subproblems of modelling DNA accessibility. For example, the DNA sequence-based deep learning models represent combinatorial interactions and in-vivo TF-DNA recognition preferences. In contrast, GRN models use TF abundance profiles across cells and in-vitro-derived TF-DNA recognition preferences, optionally incorporating ATAC-seq data as a filter. All models learn cell-type specific weights and properties and don’t generalize to new TF abundance states such as new cell types. Therefore, we are missing an end-to-end mechanistic model that represents all steps of the biological process, that generalizes to both new DNA sequences and TF abundance combinations and can simultaneously characterize hundreds to thousands of cell states observed in single-cell genomics atlases. Here, we formulated cell2state, a mechanistic end-to-end probabilistic model of TF recruitment to a chromatin locus and downstream TF effect on DNA accessibility. Cell2state is designed to achieve the generalization of regulatory predictions to unseen cell types. Cell2state A) estimates TF nuclear protein abundance and models B) how TFs recognize DNA, C) how TF sites in DNA lead to TF recruitment to a chromatin locus, D) how the activity of DNA-associated TFs affects chromatin accessibility. To evaluate generalization, we defined the computational problem and developed a workflow for predicting the scATAC-seq readout for previously unseen chromosomes and cell types. We show that cell2state outperforms the state-of-the-art deep learning models (ChromDragoNN) at explaining DNA accessibility differences across cells. Finally, to look at cell state plasticity, we developed ways to use cell2state to simulate the possible chromatin states given TF abundance of source cell types.

Speaker:  Vitalii Kleshchevnikov, PhD
Affiliation:  Wellcome Sanger Institute
Position:  Bioinformatician @ Bayraktar, Stegle, Teichmann group
Host: Daniel MacDonald, Gibson Lab

Date: Monday February 26, 2024
Time: 10:00AM-11:00AM ET
Zoom: https://partners.zoom.us/j/82163676866
Meeting ID: 821 6367 6866

Vitalii Kleshchevnikov is driven by a deep interest in three key areas: i) understanding the regulatory code which allows a single genome to specify the full diversity of cell populations and their interaction, ii) formalizing the biology of these processes into mechanistic AI/ML models, and iii) accelerating the therapy development to address ageing alterations in these processes. Vitalii did his PhD jointly supervised by Dr Omer Bayraktar, Dr Oliver Stegle, Dr Sarah Teichmann at Wellcome Sanger Institute (2018-2023) and will present the published and ongoing work. Prior to PhD, Vitalii worked on the role of peptide motifs (SLiMs) in intracellular signaling (Dr Evangelia Petsalaki, EMBL-EBI), predicting CRISR KO mutational outcomes (Dr Leopold Parts, Wellcome Sanger Institute) and profiling protein interactions in accelerated ageing (A*STAR) – while completing MSc and BSc in Kyiv, Ukraine.

Click here to be added to our mail list.