Research

Michelle M. Li, Yepeng Huang, Marissa Sumathipala, Man Qing Liang, Alberto Valdeolivas, Ashwin Ananthakrishnan, Katherine Liao, Daniel Marbach, Marinka Zitnik
Nature Methods 2024

PINNACLE: Contextual AI Model for Single-Cell Protein Biology

Understanding protein function and discovering molecular therapies require deciphering the cell types in which proteins act as well as the interactions between proteins. However, modeling protein interactions across diverse biological contexts, such as tissues and cell types, remains a significant challenge for existing algorithms. We introduce PINNACLE, a flexible geometric deep learning approach that is trained on contextualized protein interaction networks to generate context-aware protein representations. Leveraging a human multi-organ single-cell transcriptomic atlas, PINNACLE provides 394,760 protein representations split across 156 cell type contexts from 24 tissues and organs. PINNACLE’s contextualized representations of proteins reflect cellular and tissue organization. Pretrained PINNACLE protein representations can be adapted for downstream tasks: to enhance 3D structure-based protein representations at cellular resolution and to study the genomic effects of drugs across cellular contexts. PINNACLE outperforms state-of-the-art, yet context-free, models in nominating therapeutic targets for rheumatoid arthritis and inflammatory bowel diseases, and can pinpoint cell type contexts that are more predictive of therapeutic targets than context-free models. PINNACLE is a network-based contextual AI model that dynamically adjusts its outputs based on biological contexts in which it operates. Read more

Links: Paper, GitHub Repository, Interactive Demo, Data, Model Checkpoints
Press/Highlights: Cell Systems, HMS 2024 State of the School Address (skip to 35:57), Nature Methods Editorial, HMS News, Deeper Learning (Kempner Institute), Bio-IT World
Related Co-authored Papers:
* Contextual AI models for context-specific prediction in biology (Nature Methods Research Briefing, 2024)
* Signals in the Cells: Multimodal and Contextualized Machine Learning Foundations for Therapeutics (NeurIPS Workshop on AI for New Drug Modalities, 2024; Selected for spotlight presentation)
* Multi-Scale Graph Neural Network for Alzheimer’s Disease (Machine Learning for Health Findings Track, 2024)
* Deep Contextual Learners for Protein Networks (ICML Workshop on Computational Biology, 2021; Selected for oral spotlight presentation and received Best Poster Award)

Michelle M. Li, Kevin Li, Yasha Ektefaie, Shvat Messica, Marinka Zitnik
In Review 2025

CLEF: Controllable Sequence Editing for Counterfactual Generation

Counterfactual thinking is a fundamental objective in biology and medicine. “What if” scenarios are critical for reasoning about the underlying mechanisms of cells, patients, diseases, and drugs: “What if we treat the cells with the drug every one or 24 hours?” and “What if we perform the surgery on the patient today or next year?” We should reason about both the choice and timing of the counterfactual condition. Thus, counterfactual generation requires precise and context-specific edits that adhere to temporal and structural constraints. Sequence models generate counterfactuals by modifying parts of a sequence based on a given condition, enabling reasoning about “what if” scenarios. While these models excel at conditional generation, they lack fine-grained control over when and where edits occur. We develop CLEF, a controllable sequence editing approach for instance-wise counterfactual generation. CLEF learns temporal concepts that represent the trajectories of the sequences to enable accurate counterfactual generation guided by a given condition. We demonstrate that CLEF achieves state-of-the-art performance on four novel benchmark datasets in cellular reprogramming and patient immune dynamics. Read more

Links: Paper, GitHub Repository

Emily Alsentzer*, Michelle M. Li*, Shilpa N. Kobren, Undiagnosed Diseases Network, Isaac S. Kohane, Marinka Zitnik
In Review 2024
* Equal contribution (alphabetical)

SHEPHERD: Deep Learning for Diagnosing Patients with Rare Genetic Diseases

There are over 7,000 unique rare diseases, some of which affecting 3,500 or fewer patients in the US. Due to clinicians’ limited experience with such diseases and the considerable heterogeneity of their clinical presentations, many patients with rare genetic diseases remain undiagnosed. While artificial intelligence has demonstrated success in assisting diagnosis, its success is usually contingent on the availability of large annotated datasets. Here, we present SHEPHERD, a deep learning approach for multi-faceted rare disease diagnosis. To overcome the limitations of supervised learning, SHEPHERD performs label-efficient training by (1) training exclusively on simulated rare disease patients without the use of any real labeled data and (2) incorporating external knowledge of known phenotype, gene and disease associations via knowledge-guided deep learning. Read more

Links: Paper, GitHub Repository, Interactive Demo, Data, Model Checkpoints
Awards: Best Oral Presentation (ISMB 2023 TransMed; selected from 130 abstracts for long talk)
Press: Washington Post (Opinion)
Related Co-authored Papers:
* Simulation of undiagnosed patients with novel genetic conditions (Nature Communications, 2023; GitHub Repository, Data)
* Subgraph Neural Networks (NeurIPS, 2020; Project Website, GitHub Repository, Data)

Michelle M. Li, Kexin Huang, Marinka Zitnik
Nature Biomedical Engineering 2022

Perspective: Graph Representation Learning in Biomedicine and Healthcare

Biomedical networks are universal descriptors of systems of interacting elements, from protein interactions to disease networks, all the way to healthcare systems and scientific knowledge. With the remarkable success of representation learning in providing powerful predictions and insights, we have witnessed a rapid expansion of representation learning techniques into modeling, analyzing, and learning with such networks. Here, we put forward an observation that long-standing principles of networks in biology and medicine—while often unspoken in ML research—can provide the conceptual grounding for representation learning, explain its current successes and limitations, and inform future advances. Read more

Links: ISMB 2022 Tutorial Materials
Related Co-authored Papers:
* Graph Artificial Intelligence in Medicine (Annual Review of Biomedical Data Science, 2024)
* Current and future directions in network biology (Bioinformatics Advances, 2024)