Decoding Cancer From Big Data

StatisticsMachine learningData scienceAICancer biology

About

Statistical and machine learning methods that turn large-scale single-cell, spatial and multimodal cancer data into reproducible biological discoveries and clinically actionable insights.

Cancer is a spatially organized ecosystem in which tumor initiation, evolution, metastasis and treatment response emerge from the coordinated behavior of malignant and non-malignant cells. My research moves from statistical rigor to biological discovery to clinical translation: (1) developing principled methods that separate biological signal from unwanted variation, building on the RUV framework (RUV-III, PRPS, RUVprps, fastRUV-III); (2) decoding how cell states, cell–cell interactions, spatial niches and multicellular programs organize and reorganize during metastasis, organ colonization and therapy; and (3) translating these high-resolution discoveries into biomarkers and predictive models that work on routine H&E histopathology and bulk transcriptomics.

Google Scholar GitHub LinkedIn
1,255
Citations
16
h-index
$2.4M
Grant funding
10+
Trainees mentored

Experience

Appointments and education

2025 – present
Boston, USA

Senior Research Fellow, Computational Cancer Biology

Department of Medical Oncology, Dana-Farber Cancer Institute · Harvard Medical School · P.I. Prof. Himisha Beltran
Pan-cancer landscapes of transcriptional heterogeneity in metastatic tumours.
2019 – 2024
Melbourne, AU

Senior Research Fellow / Research Fellow

Bioinformatics Division, Walter and Eliza Hall Institute (WEHI) · P.I. Prof. Terence P. Speed, Prof. Anthony T. Papenfuss
Computational and statistical tools for bulk and single-cell RNA-seq, methylation, CITE-seq and spatial transcriptomics.
2014 – 2015
Melbourne, AU

Research Assistant

Translational Genomics and Epigenomics Lab, Olivia Newton-John Cancer Research Institute (Ludwig Institute) · P.I. Assoc. Prof. Alexander Dobrovic
RNA-seq and methylation analysis for predictive marker identification in lung cancer.
2014 – 2018
Melbourne, AU

PhD, Computational Cancer Biology and Bioinformatics

The University of Melbourne · Advisors Prof. Terence P. Speed, Assoc. Prof. Alexander Dobrovic
Removing unwanted variation in large and complex gene expression datasets: novel normalization methods for NanoString nCounter and RNA-seq data.
2010 – 2013
Iran

MSc, Human Molecular Genetics

Chamran University & National Institute of Genetic Engineering and Biotechnology (NIGEB)
A panel of cancer testis antigens and clinical risk factors to predict metastasis in colorectal cancer.
2006 – 2010
Iran

BSc, Human Molecular Genetics

Shahrekord University
A comparative study of amino acid composition and codon usage of Otx and Oct gene families using a bioinformatics approach.

Research

Three interconnected aims, built on the STAMP atlas

Aim 1 · Methods

Robust & reproducible cancer genomics

Principled, scalable statistical methods that distinguish biological signal from unwanted technical and confounding variation across studies, platforms and modalities.

Details
  • The RUV framework — RUV-III for NanoString nCounter (Nucleic Acids Res 2019), pseudo-replicates of pseudo-samples (PRPS) for large-scale RNA-seq (Nat Biotechnol 2023), and RUVprps, which identifies unwanted variation without prior knowledge across TCGA, CCLE, GTEx and Treehouse.
  • Atlas-scale single-cell & spatial — RUV-III-NB for single-cell counts and fastRUV-III for single-cell, CITE-seq, ATAC-seq and spatial data, normalizing millions of cells in minutes.
  • Next steps — cell-type- and spatially aware models, joint normalization of multimodal (RNA + ADT) data, and count-aware estimators that scale beyond 10 million cells.
  • Digital pathology — removing staining-center and scanner effects from foundation-model image embeddings to improve model transportability across institutions.
  • Confounding & diagnostics — identifiability theory, sensitivity analyses and RLE-based diagnostics for when biology and unwanted variation are intertwined.
Aim 2 · Discovery

Cellular & spatial organization of cancer ecosystems

Scalable, interpretable methods that resolve cell states, spatial niches and multicellular programs across millions of cells and thousands of patients.

Details
  • Reproducible cell states — reference-free annotation with ClustoCell, evaluated on ~24 million cells, extended with hierarchical models, calibrated uncertainty and detection of novel or transitional states.
  • Multicellular modules — composition-aware count models that bring the RUV principle to cell-state abundances, separating true cellular coordination from study, platform and dissociation artifacts.
  • Ecosystem rewiring — how ecosystems reorganize across normal tissue, primary tumors, metastatic sites and treatment, separating tumor-intrinsic programs from host-organ signal.
  • Tissue-of-sampling confounding — a framework showing that host-tissue signal can masquerade as metastatic tumor biology in bulk transcriptomes.
Aim 3 · Translation

Clinically accessible biomarkers & prediction

Transferring discoveries from single-cell and spatial profiling to routine H&E histopathology and bulk transcriptomics to predict progression and treatment response.

Details
  • Spatial biology from H&E — attention-based multiple-instance learning on pathology foundation-model embeddings to predict spatial ecotypes and cell states from routine slides.
  • Bulk RNA-seq — uncertainty-aware deconvolution and cross-modal transfer learning to estimate atlas-defined cell states, modules and host-tissue contributions.
  • Clinical trials — multimodal immune profiling (spatial, CITE-seq, IMC, methylation, PET) of response to 177Lu-PSMA, pembrolizumab and olaparib in metastatic prostate cancer.
  • Consortia — integration of ~2,000 prostate cancer transcriptomes from eight countries for the Pan Prostate Cancer Group.

Current work

Active projects

Active

Single-cell atlas of metastatic and primary tumours

A single-cell (18 million cells), bulk (70k samples) and spatial (2,400 samples) transcriptomic database of primary and metastatic cancers, built to resolve cell identity and state across cancer types.

Dana-Farber Cancer Institute · Beltran lab
Active

Pan-cancer transcriptional heterogeneity in metastatic tumours

Shared and context-specific molecular programs across normal, primary and metastatic disease, including removal of host-tissue background signal (RUVprps_Met) and the effect of non-malignant cells on transcriptomic analyses.

Dana-Farber Cancer Institute · Beltran lab · RUVprps_Met
Active

Joint normalization of single-cell multimodal and spatial data

Scalable normalization of single-cell multimodal and spatial data with the fastRUV-III algorithm.

RUVprps · fastRUV-III
Active

Immune landscape of metastatic prostate cancer – PRINCE and LuParp clinical trials

Immune, imaging and genomic characterization of response to 177Lu-PSMA, pembrolizumab and olaparib in metastatic castration-resistant prostate cancer.

PCF Challenge Award · PMCC & WEHI
Active

Prostate cancer consortium studies

Analysis contributions to studies of clonal dynamics and heterogeneous trajectories towards metastatic prostate cancer.

Pan-Prostate Cancer Group · panprostate.org
Active

Digital pathology for prostate cancer

Deep learning that links histopathology with molecular data, including detection of neuroendocrine prostate cancer from routine histology.

Dana-Farber Cancer Institute
Active

Removal of unwanted variation from digital pathology data

Dana-Farber Cancer Institute

Publications

Selected · † first or senior author

Nature
2026 · Nature
Integrated signatures define mutational processes in prostate cancer
Gruber A.J., …, Molania R., …, Weischenfeldt J.
Nature Communications
2026 · Nat Commun
Recurrent mechanisms of biallelic epigenetic inactivation reveal new putative tumour suppressor genes in prostate cancer
Kiriy D., …, Molania R., …, Weischenfeldt J.
Cancer Research
2026 · Cancer Res
Pan-cancer landscapes of transcriptional heterogeneity in metastatic tumors
Molania R.†, Beltran H.
bioRxiv
2026 · bioRxiv · under revision
RUVprps enables accurate and scalable RNA-seq normalization without prior knowledge of biological and unwanted variation
Trussart M., …, Molania R.†
bioRxiv
2026 · bioRxiv · submitted
ClustoCell reveals cell states and their markers from single-cell transcriptome
Salavaty A., …, Molania R.†
Cancer Research Communications
2026 · Cancer Res Commun
Vitronectin enrichment in prostate cancer liver metastases promotes adhesion and survival
Egelberg J., …, Molania R., …, Beltran H.
Nature Biotechnology
2023 · Nat Biotechnol
Removing unwanted variation from large-scale RNA sequencing data with PRPS
Molania R.†, …, Speed T.P.
Nature Biotechnology
2023 · Nat Biotechnol · briefing
Normalizing cancer RNA-seq data for library size, tumor purity and batch effects
Speed T.P., Molania R.
Nucleic Acids Research
2022 · Nucleic Acids Res
RUV-III-NB: normalization of single cell RNA-seq data
Salim A., Molania R., …, Speed T.P.
Cancer Immunology Research
2021 · Cancer Immunol Res
The ratio of exhausted to resident infiltrating lymphocytes is prognostic for colorectal cancer patient outcome
Foroutan M., Molania R., …, Huntington N.D.
Genome Biology
2021 · Genome Biol
tidybulk: an R tidy framework for modular transcriptomic data analysis
Mangiola S., Molania R., …, Papenfuss A.T.
Nucleic Acids Research
2019 · Nucleic Acids Res
A new normalization for Nanostring gene expression data
Molania R.†, …, Speed T.P.

Full list on Google Scholar →

Recognition

Funding, honours and service

Grants

  • 2024–27PCF Challenge Award US$1M
    Co-Investigator · PMCC & WEHI
  • 2021–23Merck Oncology Translational Grant US$1M
    Co-Investigator · PMCC & WEHI
  • 2021–24PCF Young Investigator Award US$225K
    Principal Investigator · WEHI & PMCC
  • 2021–23MACH Grant A$200K
    Co-Investigator · WEHI

Consortia & leadership

  • 2020–Massive Analysis and Quality Control (MAQC) Society
  • 2026–Metastatic Research Society (MRS)
  • 2025–Human Tumor Atlas Network (HTAN)
  • 2025–26MAQC Machine Learning Working Group
  • 2020–Pan-Prostate Cancer Genome Consortium (PPCG)
  • 2020–International Society for Computational Biology (ISCB)
  • 2016–Australian Bioinformatics and Computational Biology Society (ABACBS)
  • 2024–Co-Leader, PCF Microenvironment and Immunology Working Group
  • 2022Sole Convenor, Victorian Cancer Bioinformatics Symposium (VCBS)
  • 2023Co-Convener, WEHI Bioinformatics Seminar Series
    45 weekly talks

Honours & awards

  • 2021, 25Young Scientist Award
    MAQC Society, awarded twice
  • 2021Professional Award
    Post-doctoral Association, WEHI
  • 2021Travel Scholarship
    MAQC Society
  • 2018Top PhD Student Medal
    Olivia Newton-John Cancer Research Institute and The University of Melbourne
  • 2017PhD Top-Up and Travel Scholarships
    Cancer Therapeutics CRC · ONJCRI
  • 2016Best Talk Award
    COMBINE Conference, Brisbane
  • 2014Melbourne International Research & Fee Remission Scholarships for PhD
    The University of Melbourne

Mentorship

  • PostdocsFour research fellows at WEHI and PMCC
  • StudentsSix Master's students across DFCI, WEHI, The University of Melbourne and Monash University
  • PhDPhD advisory committee member
    WEHI / The University of Melbourne · 2022–2025

Review

  • JournalsNature Communications, Nucleic Acids Research, NAR Genomics and Bioinformatics, Scientific Reports, BMC Cancer, BMC Research Notes, Computer Methods and Programs in Biomedicine, Discover Oncology

Talks & teaching

† invited speaker

Selected talks

  • 2025Pan-cancer landscapes of transcriptional heterogeneity in metastatic tumors
    MAQC 2025, SAS Institute, Cary, NC
  • 2025Fast, accurate and scalable normalization for RNA sequencing data with the RUVprps software package
    MAQC 2025, SAS Institute, Cary, NC
  • 2024Fast and accurate normalization of atlas-level single cell data †
    Bioinformatics Platform, Monash University
  • 2023Removing unwanted variation from large cancer gene expression data †
    International Biometric Society, Eastern Mediterranean Region, Turkey
  • 2022RUV-III-PRPS for normalizing large and complex RNA-seq data †
    Statistical Methods in Genetic/Genomic Studies Workshop, National University of Singapore
  • 2022Normalizing RNA-seq data for library size, purity and batch effects †
    MAQC China Chapter Meeting
  • 2018A new normalization for Nanostring gene expression assay
    useR! Conference, Brisbane
  • 2016Robust normalization of Nanostring gene expression data
    ABACBS / COMBINE Conference, Brisbane

Teaching & workshops

  • 2023Workshop instructor: Removing unwanted variation from large cancer data
    International Biometric Society Conference, Turkey
  • 2022Designer & instructor: Experimental design course (graduate)
    WEHI Statistics Education Subcommittee
  • 2018Workshop co-organizer & instructor: Removing unwanted variation with R
    useR! 2018, Brisbane
  • 2016Teaching assistant: Data visualization in R
    Melbourne Business School, University of Melbourne
  • 2016Guest lecturer: Algorithms for functional genomics
    The University of Melbourne
  • 2016Guest lecturer: Gene expression data analysis: pitfalls and principles
    The University of Melbourne

Software & resources

Open source

Atlas · single-cell, bulk & spatial

STAMP

Single-cell Transcriptomic Atlas of Metastatic and Primary tumors: ~18 million single cells from 5,500 samples, 70,000 bulk transcriptomes and 54 million cells and spots from 2,400 spatial transcriptomic samples, uniformly processed, quality-controlled and normalized.

Details
  • Status — manuscript in final preparation; public release planned through an interactive web resource.
  • Single-cell — computational pipelines for cell-type and cell-state annotation across cancer types and disease sites.
  • Bulk — major sources of unwanted variation identified, molecular subtypes characterized, and cellular composition estimated using the single-cell atlas as a reference.
  • Spatial & H&E — matched spatial profiling and histopathology for studying tissue organization and training image-based predictors.
  • Growing — an automated search and data-ingestion pipeline brings in newly released datasets, with priority on multimodal, emerging single-cell and spatial technologies.
  • Purpose — a resource for studying cancer evolution across primary and metastatic disease, and a demanding test bed for new statistical methods.
  • Developed at DFCI (Beltran lab), 2025–2026. Preliminary results presented at AACR 2026.
Website · RUV hub

RUVtools

A unified resource for RUV methods, software, documentation and reproducible vignettes.

Details
  • Status — under development.
  • Purpose — one place to find every RUV normalization tool and how to use it.
  • Content — detailed vignettes and comprehensive documentation.
  • Collaborators — Prof. Terence P. Speed and Assoc. Prof. Johann Gagnon-Bartsch (University of Michigan).
  • Developed at WEHI and DFCI, 2023–2026.
Website
R package · bulk RNA-seq

RUVprps

Accurate and scalable normalization and integration of large and complex RNA-seq data.

Details
  • Status — completed; preprint under revision at Nucleic Acids Research.
  • Method — identifies sources of unwanted variation directly from the data and constructs pseudo-replicates of pseudo-samples (PRPS) without prior knowledge of biological or unwanted variation.
  • Validated on TCGA, CCLE, GTEx and Treehouse.
  • Scope — more than 100 functions for assessing, removing and diagnosing unwanted variation.
  • Developed at WEHI and DFCI, 2024–2026.
GitHubPreprint
R package · single-cell & spatial

fastRUV-III

Fast normalization and integration of atlas-scale single-cell, multimodal and spatial data.

Details
  • Status — completed; manuscript in final preparation.
  • Data — single-cell RNA-seq, CITE-seq (ADT), ATAC-seq, and imaging- and sequencing-based spatial assays.
  • Speed — normalizes 5 million cells across 8,000 genes in under five minutes on a single CPU core.
  • Benchmarks — performs strongly against scVI, Harmony and BBKNN.
  • Related — RUV-III-NB for single-cell counts (Nucleic Acids Res 2022).
  • Developed at WEHI and DFCI, 2024–2026.
GitHub
Method · single-cell

ClustoCell

Reference-free identification of cell identities, cell states and their markers from single-cell transcriptomes.

Details
  • Status — preprint, manuscript under review.
  • Approach — stratifies expression within each cell into high- and intermediate-expression tiers to build similarity graphs driven by intrinsic expression programs.
  • Scale — evaluated across ~24 million cells spanning multiple tissues, diseases and platforms.
  • Role — senior author; development supervised at PMCC and DFCI.
Preprint
R package · metastasis

RUVprps_Met

Removal of host-tissue background signal from metastatic bulk RNA-seq.

Details
  • Status — completed; manuscript in final preparation.
  • Problem — the tissue a metastasis is sampled from (e.g. bone) can contribute strong host-tissue programs that are mistaken for tumour biology.
  • Approach — uses STAMP single-cell and spatial data to estimate host-tissue, immune and stromal contributions, and removes them within the RUVprps framework.
  • Developed at DFCI, 2026.
GitHub
R package · TCGA

tcgaCleaner

Exploring and removing unwanted variation from TCGA RNA-seq data.

Details
  • Status — completed.
  • Use — assess library size, batch and tumour-purity effects in TCGA RNA-seq and remove them.
  • Developed at WEHI, 2022.
Web app · TCGA

TCGA R Shiny app

Interactive visualization for exploring and removing unwanted variation from TCGA pan-cancer RNA-seq data.

Details
  • Status — completed.
  • Use — explore sources of unwanted variation interactively across the TCGA pan-cancer data.
  • Built with R Shiny.
  • Developed at WEHI, 2022.
Open app
Pipeline · consortium

PPCG RNA-seq

Integration and normalization of ~2,000 prostate cancer transcriptomes from eight countries for the Pan Prostate Cancer Group.

Details
  • Status — ongoing.
  • Role — led integration and normalization for the PPCG RNA Working Group, co-led by Assoc. Prof. Luigi Marchionni and Prof. Anthony T. Papenfuss.
  • Outcome — RUV-adjusted transcriptomic data used across consortium analyses, including the pan-prostate cancer genome study in Nature.
  • Involved since 2021.
Overview
Data resource · 7k downloads

TCGA pan-cancer RNA-seq

Harmonized TCGA RNA-seq data for 33 cancer types and about 11,000 samples, as SummarizedExperiment objects with raw counts, FPKM, batch information and gene annotations.

Details
  • Status — completed.
  • Coverage — 33 cancer types and about 11,000 samples.
  • Contents — raw counts, FPKM, batch information and gene annotations.
  • Format — SummarizedExperiment objects, ready for R and Bioconductor workflows.
  • Reach — 7k downloads; released from WEHI, 2023.
Zenodo

Technical skills

Methods, data and tools

Statistics

Statistical methodology

  • Removing unwanted variation (RUV-III, PRPS)
  • Normalization & batch correction
  • Experimental design
  • Negative-binomial & generalized linear models
  • Factor analysis & low-rank methods
  • Hierarchical & compositional models
  • Confounding & sensitivity analysis
  • Benchmarking & simulation
Machine learning

Machine learning & AI

  • Deep learning (PyTorch)
  • Computational pathology
  • Pathology foundation models (UNI)
  • Attention-based multiple-instance learning
  • Cell-type & cell-state annotation
  • Clustering & graph methods
  • Deconvolution
  • Predictive modelling
Data

Data types

  • Bulk RNA-seq
  • Single-cell RNA-seq
  • CITE-seq
  • ATAC-seq
  • Spatial transcriptomics
  • NanoString nCounter & GeoMx DSP
  • Imaging mass cytometry
  • DNA methylation
  • Exome & genome sequencing
  • TCR/BCR
  • Whole-slide H&E images
Computing

Programming & infrastructure

  • R & Bioconductor
  • R package development
  • R Shiny
  • Python
  • Bash & Linux
  • HPC (SLURM)
  • Git & GitHub
  • Docker & Singularity
  • Nextflow & Snakemake
  • Atlas-scale data pipelines

Honorary appointments

Affiliated groups

Honorary
Melbourne, AU

Speed Lab

Walter and Eliza Hall Institute (WEHI) · Prof. Terence P. Speed
Friend of the lab
Boston, USA

Irizarry Lab

Dana-Farber Cancer Institute (DFCI) · Prof. Rafael Irizarry
Friend of the lab
Melbourne, AU

Sandhu Lab

Peter MacCallum Cancer Centre (PMCC) · Prof. Shahneen Sandhu

Contact

Collaboration, methods, software, speaking