Raúl Rabadán, Columbia University
Meet: https://meet.google.com/niy-gtpk-sro
YouTube Stream: https://youtube.com/live/i7ETtL3CwcY
Join group to receive calendar invite: https://groups.google.com/a/modelingtalks.org/g/talks
Abstract:
This talk traces the history of our understanding of cancer as a disease of our own genome, and how cancer genomics—the technologies and the data it generates—has identified key proteins driving oncogenesis, tumor evolution, and responses to therapy, including resistance. Yet protein-coding mutations account for only ~1% of all mutations found in cancer patients; the other 99%—mutations in non-coding regions—remain largely uncharacterized. I will discuss how computational approaches are beginning to map the functional roles of mutations in the non-coding cancer genome, where such alterations reshape gene expression and cell state.
Bio:
Raúl Rabadán (born 1974) is a Spanish-American theoretical physicist and computational biologist. He is currently the Gerald and Janet Carrus Professor in the Department of Systems Biology, Biomedical Informatics and Surgery at Columbia University. He is the director of the Program for Mathematical Genomics at Columbia University and previously the director of the Center for Topology of Cancer Evolution and Heterogeneity (2015-2021). He is the co-leader of the Cancer Genetics and Epigenetics Program at the Herbert Irving Comprehensive Cancer Center at Columbia University. Dr. Rabadan received the 2021 Outstanding Investigator Award by the National Cancer Institute. At Columbia, he leads a highly interdisciplinary team of researchers from the fields of mathematics, physics, computer science, engineering, and medicine, with the common goal of solving pressing biomedical problems through quantitative computational models. Rabadan's current interest focuses on uncovering patterns of evolution in biological systems—in particular, viruses and cancer.
Summary:
Focus: appearance and dynamics of cancer
History:
Cancer has been known for millenia with many diagnostic descriptions
Theodor Boveri: observed that cancer cells are genomically abnormal and based on a single cell’s mutation that gets cloned as the cell replicates
Types of carcinogens discovered:
Peyton Rous: observed propagation of tumors across animals via viruses
Katsusaburo Yamagiwa: rabbit tumors, demonstrated role of coal tar as carcinogen
Hermann Muller: showed how radiation causes cancer
Carl Nordling: vulnerability of people to cancer increases as age following the same pattern across many populations (indicates a common age-related process)
Michael Bishop and Harold Varmus: observed that viruses change the expression of genes in our own cells, led to new work on cancer generics
Peter Nowell: cancers come from single cell that was mutated; a single or series of mutations that cause the cell to replicate out of control create cancer
Genomic analysis:
Germ-line mutations: inherited from parents
Normal cells mutate on a regular basis (carcinogens make this more frequent); a few key mutations cause out of control replication: cancer
We have been developing a global map of the different types of mutations that cause cancer
Standard model of cancer evolution:
Normal cells
Random mutation -> oncogenesis (cell replicates out of control)
Diagnosis
Treatment of main tumor
A subset of cancerous cells may remain and keep replicating -> Relapse
Metastasis -> population of cells becomes resistant to treatment and/or spreads over much more of the body
We have developed a dynamic map of mutations that appear during Diagnosis and Relapse, the mutations that can be diagnosed and the mutations that tend to appear during Relapse, specifically as an evolutionary response to treatment (mutations that help cancer cells survive treatment)
Non-coding mutations in cancer
99% of mutations occur in non-coding genome regions
Gene promoters, enhancers, etc.
When we see more mutations in non-coding regions this indicates some selective pressure, which implies that they have some functional role
Different cells of origin accumulate different mutations
GET (general expression transformer): an interpretable foundation model designed to uncover regulatory grammars across 213 human fetal and adult cell types (https://www.nature.com/articles/s41586-024-08391-z)
Model expression using genomic and epigenomic data
Gene -> expression: Predict expression in a given cell type given epigenetic data
Leveraging language models using space of regulatory regions as language
g(x): Regulatory grammar
f(x): Contribution of region to different gene expression
Foundational model that was finetuned to specific cancer types
One-shot prediction: fine-tune on a single patient of a population, predict on the rest
Improves prediction accuracy for the tumor, macrophages (immune) and oligodendrocytes (brain neural cells)
Applications:
Assessing regulatory impact of non-coding mutations
GLASS: Glioma Longitudinal Analysis
64 patients, 2-6 tumor samples
Documents evolution of mutations over time
10-1k coding mutations
1k-1m non-coding mutations
Using foundation model to assess impact of non-coding mutations
Predicting expression of proteins given such mutations
Makes it possible to identify the cell type from which a given tumor cell originated based on the mutation that occurred
Interaction of combinatorial regulatory factors in a particular cell type
Looking for a cell of origin for a particular tumor
Then find the transcription factors that regulate/interact with the mutated transcription factor
Used AlphaFold to model direct protein interactions among these transcription factors
Observed that cancer mutation disrupted the transcription factor interactions
Applications of GET in cancer genomics
Identifying regulatory elements for a particular gene in particular cell type
Assessing the regulatory impact of non-coding mutations
Interactions of regulatory factor in each cell type
Directions for GET finetuning:
Brain tumors
DNA repair genes
Pancancer: 19 tumor types
MDS/AML Stem Cells
Prostate Cancer
Lymphonas