Roger Casals-Franch

PhD method · UVic-UCC

A Boolean gene regulatory network framework for mechanistic modelling of cell-state transitions. Networks inferred from single-cell multi-omic data are constrained to reproduce the cell states and transitions actually observed, so that in silico perturbations can be read mechanistically.

• Boolean networks
• GRN inference
• Perturbations

Presented at JBI 2025, scverse 2025 (Stanford) and ECCB 2026.

PhD · with L. Nonell, J. Villà-Freixa, A. López García de Lomana

Trajectory inference and gene regulatory network analysis of 25,581 tumour-infiltrating T cells from squamous cell carcinoma, before and after anti-PD-1. Resolves memory-to-exhaustion trajectories driven by EOMES and TCF7, Tpex vs Tex rewiring, and enhanced CXCL13⁺ Tfh responses.

• Trajectory inference
• SCENIC
• TCR clonotypes

Saez-Rodriguez group, Heidelberg · 2024

A modular Snakemake framework to compare methods that infer gene regulatory networks from paired and unpaired single-cell RNA + ATAC data. Inferred networks proved highly sensitive to method choices and struggled to capture causal effects from perturbation assays.

• Benchmark
• scRNA + scATAC
• Snakemake

IrsiCaixa · group of Christian Brander

Single-cell transcriptomic and epigenomic analysis of BCN03, a therapeutic HIV vaccine trial, to characterise vaccine-induced changes in gene regulation. T- and NKT-cell state annotation is integrated with TCR repertoire analysis to compare clonotype usage between people with HIV and healthy controls.

• scRNA + scATAC
• TCR repertoire
• NKT cells

with B. Casajuana, A. López García de Lomana, P. Martí-Puig, J. Villà-Freixa · 2026

How reliably can PINNs recover kinetic parameters of a synthetic genetic oscillator? Trajectories are reconstructed well, but parameter recovery is fragile under noise, sparse sampling, unobserved repressors and poor initial guesses, especially in the oscillatory regime.

• PINNs
• ODE inverse problems
• Dynamical systems

Open Problems – Single-Cell Perturbations (Kaggle)

Predicted gene-expression responses to unseen drug perturbations by combining scRNA-seq and scATAC-seq, using SCENIC+ regulatory networks as priors for machine-learning and variational autoencoder models. Ranked in the top 10% of more than 1,000 teams.

• VAEs
• SCENIC+
• Multi-omics

Workflow · Snakemake

Donor demultiplexing of pooled 10x single-cell data from raw FASTQs: Cell Ranger multi, genotype-based clustering with souporcell, HLA typing of each cluster with arcasHLA, and matching back to known donor HLA alleles.

• 10x Genomics
• HLA typing
• Snakemake

Biomedical Center, University of Iceland · 2025

Context-specific genome-scale metabolic models built from RNA-seq with rFASTCORMICS, constrained with Seahorse measurements and analysed with COBRApy, to predict metabolic responses to treatment.

• Constraint-based models
• COBRApy
• RNA-seq

Team project, Group 9 · with I. Carbó, M. Navarro, G. Ruiz · 2026

A hierarchical Bayesian model that predicts, from the 3-month follow-up visit, whether lung diffusing capacity (DLCO) will recover by 6 months and whether the patient will return, across three post-COVID and post-ICU cohorts. Validated on unseen hospitals and served through a web app.

• Bayesian models
• Clinical prediction
• Web app