Cells in healthy individuals were thought to carry the same number of chromosomes, but it was recently discovered that in the breast tissue of healthy women there are rare cell populations that acquire extra or lose copies, a condition called aneuploidy. These abnormal cells mirror the characteristics of breast cancers and may represent early precursors of the disease years before clinical diagnosis. Dr. Kim focuses on finding and characterizing these rare abnormal cells in healthy breast tissues. He hopes to build computational tools to understand what biomarkers set these cells apart from their healthy neighbors and how their surrounding microenvironment may influence their behavior, opening new avenues for early cancer detection and risk stratification. To test whether these signs truly predict cancer development, he plans to apply this approach to breast tissue samples from patients who were monitored over many years, some of whom later developed cancer, with the goal of giving clinicians better ways to detect breast cancer earlier and to identify at-risk patients.

Dr. Kim will develop a heterogeneous graph neural network (GNN) that jointly models single cell copy number alterations and gene expression with genes, cells, and chromosome segments as nodes. This will separate transcriptional changes driven by chromosomal gains and losses from other sources of variation. He will extend the model to spatial transcriptomic data to further isolate microenvironmental influences on gene expression and understand how these rare aneuploid cells interact with their local environment.

Dr. Kuthyar studies why cancer patients, especially those receiving treatments like chemotherapy or radiation, are at high risk of developing serious lung infections such as pneumonia. While these treatments are essential for killing cancer cells, they also weaken a key part of the immune system that normally helps the body detect and eliminate bacteria. This weakened defense makes patients more vulnerable to infection. At the same time, many hospitalized patients receive supplemental oxygen, which can change the lung environment in ways that help certain bacteria grow stronger and become more aggressive. In cancer patients, these two factors are closely connected: the weakened immune system cannot effectively control bacteria, while the high-oxygen environment actively promotes bacterial survival and virulence. Together, this creates a perfect storm that increases both the risk of contracting pneumonia and severity of disease. This work is relevant to cancers commonly treated with immune-suppressing therapies, including leukemia, lymphoma, and solid tumors such as lung, breast, and colorectal cancer, and aims to identify better ways to predict, prevent, and treat these life-threatening infections.

This project proposes a framework to dissect pneumonia risk in immunocompromised patients using human and mouse models. Dr. Kuthyar will use hierarchical networks to link gene expression and metabolites. Multi-omics factor analysis will capture microbial and immune variation and models trained on human data will be tested in mice, enabling iterative prediction and validation. This approach integrates species harmonization, metabolite prioritization, and network mapping to reveal hyperoxia-driven microbial adaptation and myeloid immune deficits driving pneumonia risk.

The X and Y chromosomes play a crucial role in human sex determination. Females have two copies of the X chromosome, while males have one X chromosome and one Y chromosome. In females, the second copy of the X chromosome is silenced early in development, meaning that only one of the X chromosomes is expressed. As a result, mutations on the activated X chromosome are more likely to change cellular functions and in context of cancer, could lead to more rapid disease progression. Dr. Leventhal proposes a novel computational approach to distinguish between the actively expressed and silenced X chromosomes in females. He hopes to use this method to analyze a dataset of over 8,000 tumors to identify potential new drivers and molecular vulnerabilities within 31 types of cancer. He will model whether these alterations can occur in pre-cancerous cells, indicating that they could be targets for early therapeutic intervention.

Dr. Leventhal will develop a computational tool that models the error rate of statistical phasing in bulk whole-genome sequencing and corrects these errors to determine accurate haplotype-specific copy number of all chromosomes. This model integrates genomics with RNA-seq data to determine the active and inactive X chromosome. The subsequent error correction will allow him to perform the first pan-cancer analysis in over 8546 tumors to identify recurrent copy number alterations affecting the active or inactive X chromosome.

Radiopharmaceuticals, or drugs that contain radioactive forms of chemical elements, have transformed cancer diagnosis and treatment. Radioactive copper and manganese, for example, play a crucial role in PET imaging, while radioactive lutetium is used to deliver targeted radiation to cancer cells. While these radiometals have tremendous potential, however, their application is hindered by a lack of efficient “chelators,” or molecules that can securely bind radiometals in the human body. Computational protein design offers a solution by engineering protein-based chelators optimized for radiometal coordination, stability, and biocompatibility. Using advanced protein modeling, Dr. Klauser [Marilyn and Scott Urdang Quantitative Biology Fellow] will develop chelators for radiometals, improving diagnostic imaging and advancing lutetium-based radiotherapies. While this work is applied to HER2-positive gastric cancer, these strategies have broad applications across various cancer types, ultimately enhancing precision oncology and expanding radiopharmaceutical utility.

This research develops a computational strategy to design stable, compact metal-binding proteins for radiopharmaceuticals, enabling fusion with therapeutic antibodies. Using diffusion models, such as RFdiffusion, thousands of protein backbones are generated for metals like copper and manganese. Sequences are assigned via ProteinMPNN, filtered for stability and binding with AlphaFold 3. For rare lanthanides, symmetric duplication of known binding motifs is used. This approach streamlines the discovery of stable scaffolds for radiopharmaceutical applications.

Many cancer mutations occur in regions of the human genome that do not code for proteins. These non-coding regions serve as vital regulators of gene expression; mutations in these regions contribute to various hallmarks of cancer. Elucidating these regulatory elements and their malignant variants is critical for advancing our understanding of cancer biology and fostering precision medicine. Deep learning sequence models can substantially enhance our grasp of the regulatory genome in both health and disease. To this end, Dr. Wang aims to combine generative AI models with single-molecule regulatory genomics to uncover the principles that underlie the cancer regulatory genome at unprecedented resolution and precision. 

With single-molecule regulatory genomics, Dr. Wang will develop a deep generative AI model to learn the probability landscape of the single-molecule regulatory genome. By taking any DNA sequence as input, the deep generative AI model can generate diverse configurations of single-molecule chromatin states.

Triple-negative breast cancer (TNBC) is one of the most aggressive and difficult-to-treat forms of breast cancer. Dr. Bruno’s research focuses on how changes in DNA packaging, known as chromatin, affect cancer progression and treatment response. Using advanced computational models, Dr. Bruno will investigate the interplay between chromatin modifications and resistance mechanisms, with the goal of identifying strategies to mitigate therapy resistance and reduce cancer recurrence. While Dr. Bruno’s primary focus is TNBC, this research provides a generalizable framework that could be applied to other cancers where chromatin modifications play a key role, leading to better, more durable treatments for a broader range of patients. 

In this project, Dr. Bruno will develop a computational framework that integrates mathematical models of chromatin modification dynamics with pharmacokinetic and pharmacodynamic models to study the role of these modifications in triple-negative breast cancer (TNBC) progression. Dr. Bruno will adapt existing chromatin modification circuit models to TNBC dynamics and, using Bayesian inference methods, will parameterize them with experimental data. By integrating pharmacokinetic models for specific drugs, this framework will enable the evaluation of diverse treatment strategies.

Macrophages are a major component of the body’s first line of defense, acting as sentinel cells that detect and respond to threats. One powerful trait of macrophages is their ability to modulate their response based on previous exposure to stimuli. In cancer, this adaptability can steer macrophages either to fight tumors or to protect them, depending on prior experiences. This phenomenon is often referred to as “macrophage memory.” Though macrophage memory is increasingly recognized as a factor influencing cancer progression and treatment outcomes, the mechanisms that allow macrophages to retain this memory remain unclear. Dr. Park hypothesizes that exposure of macrophages to certain stimuli leads to lasting changes in the structure of their DNA. She will combine both experimental and computational approaches to elucidate how this memory forms and how it affects the expression of immune-related genes. By uncovering the principles of macrophage memory formation, she will lay the groundwork for strategies to reprogram macrophages, potentially enhancing anti-tumor immunity and improving cancer therapy.

Dr. Park will use machine learning–aided genome modeling, leveraging bulk Hi-C data, to reconstruct 3D chromosome structures and validate them with deep learning–based image analysis. This framework will infer single-cell 3D chromosome structure in macrophages. Dr. Park will also quantify nuclear speckle–mRNA spatial relationships using microscopy and develop mathematical models of gene regulation that incorporate transcription factor activity. This integrative approach will reveal how chromosome structure shapes macrophage immune memory and functional response in cancer.

Dr. Zweig [Sijbrandij Foundation Quantitative Biology Fellow] is modeling how gene regulation changes across time and space using advanced geometry, deep learning models, and CRISPR gene-editing technology. With these tools, it is possible to learn how RNA expression evolves in a cell over time and, ideally, predict evolution in regions susceptible to cancerous mutation(s). This work has the potential to apply to many types of cancer but is especially applicable for acute myeloid leukemia (AML). This is in part because there are known “precursor” states—when cells are at higher risk of becoming cancerous—where understanding gene regulation is most impactful, and in part because a common element of AML treatment is a stem cell transplant, which may attack host cells inconsistently in different regions of the same organ.

The project models temporal gene dynamics via stochastic differential equations, parameterizing the drift vector field with linear networks or shallow neural networks to guarantee provable identifiability, trained via differentiation through the adjoint method. Spatial interactions between spot clusters are characterized with graph neural networks, symmetric functions parameterizing summary statistics, and self-attention modules on latent gene embeddings. The latent embeddings are also defined through a variational autoencoder integrating RNA with other modalities.

Dr. Roman [Leslie Cohen Seidman Quantitative Biology Fellow] aims to develop mathematical tools to determine which genes are associated with resistance to chemotherapy. Given genomic information from pancreatic cancer patients whose tumors are resistant or sensitive to chemotherapy, this tool will identify genes that distinguish the two populations. These genes can then be explored as potential drug targets that can sensitize chemotherapy-resistant tumors to treatment.

Dr. Roman’s research relies on the use of information theory to improve the ability of neural networks to find genes whose RNA expression distinguishes chemotherapy-sensitive from resistant patients. Another research direction is to leverage prior knowledge, accumulated over decades about gene-gene interactions in the laboratory, to inform the architecture of the neural networks or use large foundation models training on millions of cells to study cancer.

Chromatin remodelers are complex protein machines responsible for packaging DNA and regulating gene expression. Their dysfunction is strongly implicated in cancer. For example, certain types of sarcoma and ovarian cancer are driven by mutations in a chromatin remodeler called BAF. Combining experiments with theoretical work, Dr. Owen’s research aims to understand how remodelers recognize their target sites in the cell’s nucleus. By expanding our understanding of chromatin remodeling, the findings of this research will provide the groundwork for more effective cancer treatments—suggesting how drugs might target chromatin remodelers—as well as enhance our understanding of how existing drugs that target remodeler-adjacent mechanisms might work.

A central aim of this project is the development of new, quantitative models to explain the behavior of chromatin remodelers seen in experiments. Dr. Owen will achieve this by successive rounds of passing between theory and experiments repeatedly—measuring, modeling, then measuring again. For comparison to experiments, model predictions will be extracted computationally (e.g., numerically solving ODEs, or by exact stochastic simulation using Gillespie’s algorithm) or analytically (e.g., by the King-Altman procedure, and variants), as appropriate.

Only 3% of cancer drugs in clinical trials ultimately receive FDA approval, compared to 15-33% of drugs for other types of diseases. Recent studies have suggested that many drugs being explored for cancer treatment do not actually target their intended molecule in the cell. This has important implications for efficacy and safety and could be a key contributor to the low FDA approval rate. Dr. Grabski [Kenneth G. Langone Quantitative Biology Fellow] has created a novel experimental and computational framework to identify drug mechanisms of action at molecular resolution by leveraging CRISPR-based technologies. With this framework, she hopes to more precisely identify how a given cancer drug functions in the cell. This could serve as a powerful tool for preclinical evaluation and even potential discovery of new cancer therapeutics.

Dr. Grabski’s project aims to identify drug targets by modeling drug transcriptional response as a sum of genetic perturbation responses. She will perform this deconvolution in two steps. First, she will use a multi-condition latent factor model to produce denoised estimates of perturbation effects. Second, she will leverage sparse Bayesian regression techniques to map drug responses to these perturbation effects, in a way that can summarize complex patterns of uncertainty among related perturbations.