Mrinmoy Saha Roddur

Mrinmoy Saha Roddur

Computational biologist. I reconstruct how tumors evolve, diversify, and spread — from bulk, long-read, and single-cell sequencing data.

On the job market. Available from December 2026 for research faculty and industry research positions.

About

I work on how tumors evolve — how they restructure their genomes and amplify oncogenes, diversify into genetically distinct subpopulations within a single patient, and spread from one organ to another. None of it is observed directly. It has to be reconstructed from sequencing data, and I build the methods that do the reconstructing.

My current focus is focal oncogene amplification: extrachromosomal DNA and breakage–fusion–bridge cycles, two of the most aggressive routes to drug resistance. I work across bulk whole-genome, long-read, and single-cell sequencing, on patient cohorts in lung, prostate, breast, and ovarian cancer and in acute myeloid leukemia.

PhD candidate at the University of Illinois Urbana-Champaign, advised by Mohammed El-Kebir. Previously B.Sc. at the Bangladesh University of Engineering and Technology, and lecturer at Southeast University, Dhaka.

Research

ecDNA unequal segregation BFB cycle break fuse amplify
Two routes to focal oncogene amplification: circular ecDNA that segregates unequally at mitosis, and breakage–fusion–bridge cycles that build inverted repeats on the chromosome.

Focal oncogene amplificationCurrent

Some of the most aggressive tumors amplify oncogenes not by duplicating them in place, but by restructuring the genome — spinning copies off onto circular extrachromosomal DNA, or repeatedly breaking and refusing chromosome ends in breakage–fusion–bridge cycles. Because ecDNA carries no centromere, it segregates randomly at mitosis, letting a tumor swing its oncogene dosage far faster than chromosomal inheritance allows — much of why these tumors escape treatment.

Both mechanisms leave repetitive signatures that conventional copy-number callers cannot resolve. I develop methods to reconstruct them from bulk and single-cell DNA sequencing, trace how they evolve under therapy, and quantify their contribution to cancer cell fitness.

primary met 1 met 2 migration
A clone tree annotated with anatomical sites. Dashed edges are migrations; many distinct labelings can explain the same tree equally parsimoniously.

Migration histories of metastatic cancers

Metastasis is what makes cancer lethal, but the route a tumor took through the body — which site seeded which, and in what order — is never observed. It has to be read back out of the clonal phylogeny, and whether metastases seed each other or all descend from the primary changes how the disease is understood and treated.

Existing methods report one "most likely" route, hiding how much ambiguity the data actually leaves. MACH2 recovers every migration history the data supports equally well. It outperforms existing methods on simulated and real lung, prostate, breast, and ovarian cancer cohorts. Published in Nature Methods.

IgM IgG IgA hypermutation
A B cell lineage tree. Node color is the antibody isotype, changing at class switch events; diamonds along branches are somatic hypermutations.

B cell clonal lineage trees

An antibody response is evolution compressed into weeks: B cells mutate their antibody genes, get selected on binding affinity, and switch isotype to change the antibody's function. Reconstructing that history matters for vaccine response, autoimmunity, and B cell malignancies.

Most lineage methods read only the mutations and discard the isotype switches. TRIBAL uses both, from single-cell RNA sequencing data, recovering the expected trends of affinity maturation in a model system. Published in Cell Genomics.

Data & Expertise

Sequencing data

  • Bulk whole-genome & whole-exome
  • Long-read sequencing
  • Single-cell DNA (MissionBio Tapestri)
  • Single-cell RNA
  • B cell receptor repertoires

Biological domains

  • Intratumoral heterogeneity
  • Clonal architecture & tumor phylogenetics
  • Focal amplification: ecDNA, BFB cycles
  • Copy-number & structural variation
  • Metastatic dissemination
  • Antibody affinity maturation

Tooling

  • Python, R, C++, Bash
  • PyTorch, scikit-learn, pandas
  • Nextflow, Snakemake, Git
  • Docker, Linux / HPC clusters
  • SAMtools, IGV
  • Gurobi, NetworkX

Software

MACH2

Python, Gurobi

Enumerates the full space of optimal migration histories for metastatic cancers via integer linear programming.

TRIBAL

Python

Isotype-aware reconstruction of B cell clonal lineage trees from single-cell sequencing data.

Publications

Selected

M.S. Roddur, V. Ramavarapu, A. Bunkum, A. Huebner, R. Minyev, N. McGranahan, S. Zaccaria, and M. El-Kebir. Characterizing the Solution Space of Migration Histories of Metastatic Cancers with MACH2.
RECOMB 2025Nature Methods (in press)
L.L. Weber, D. Reiman, M.S. Roddur, Y. Qi, M. El-Kebir, and A.A. Khan. Isotype-aware Inference of B cell Clonal Lineage Trees from Single-cell Sequencing Data.
Cell Genomics 4.9 (2024)ext. of RECOMB 2024
M.S. Roddur, S. Snir, and M. El-Kebir. Enforcing Temporal Consistency in Migration History Inference.
J. Computational Biology 31(5), 396–415ext. of WABI 2023
L.L. Weber, D. Reiman, M.S. Roddur, Y. Qi, M. El-Kebir, and A.A. Khan. TRIBAL: Tree Inference of B Cell Clonal Lineages.
RECOMB 2024
M.S. Roddur, S. Snir, and M. El-Kebir. Inferring Temporally Consistent Migration Histories.
WABI 2023

Earlier work

J. Willson, M.S. Roddur, B. Liu, P. Zaharias, and T. Warnow. DISCO: Species Tree Inference Using Multi-Copy Gene Family Tree Decomposition.
Systematic Biology 71.3 (2022), 610–629
J. Willson, M.S. Roddur, and T. Warnow. Comparing Methods for Species Tree Estimation With Gene Duplication and Loss.
AlCoB 2021

Teaching

Graduate Teaching Assistant, University of Illinois Urbana-Champaign
Computational Cancer Genomics · Data Structures
Lecturer, Southeast University, Dhaka  2019–2020
Data Structures · Algorithms · Machine Learning · Probability