DECODE

DECODE develops Transformer-based foundation models that learn the language of gene regulation directly from single-cell data, integrating single-cell transcriptomics (scRNA-seq) and chromatin accessibility (scATAC-seq) to infer Gene Regulatory Networks: the mechanisms that govern how cells develop and how disease progresses.

Aims

Single-cell sequencing can now profile the transcriptome and the chromatin landscape of millions of individual cells, but turning those measurements into a mechanistic account of regulation (i.e., which genes control which, in which cell type, and under which conditions) remains an open problem. DECODE addresses it by treating gene expression as a language and learning its grammar at scale.

The project pursues three objectives:

• Build a foundation model for single-cell transcriptomics that learns transferable representations of cells and genes from large, heterogeneous public data.
• Integrate transcriptomic and chromatin-accessibility data so that regulatory relationships are inferred from complementary evidence rather than expression correlation alone.
• Derive Gene Regulatory Networks that are cell-type-specific, testable against experimental evidence, and useful for studying cell development and disease progression.

Expected results

• A pre-trained single-cell foundation model with representations that transfer across tissues and datasets, evaluated on both in-distribution accuracy and cross-tissue generalisation.
• A multimodal extension that couples expression with chromatin accessibility for regulatory inference.
• Zero-shot Gene Regulatory Networks benchmarked against curated regulons and experimental evidence, and compared with established network-inference methods.
• Methodological contributions on scaling behaviour, pre-training objectives and evaluation practice for single-cell foundation models.

Results achieved

This section will be updated as the project progresses, reporting the results obtained and the outputs produced with the support of the funding received.

Funding

This project is funded by the Italian Ministry of University and Research (MUR) under the Fondo Italiano per la Scienza (FIS) — Starting Grant, with Andrea Tangherloni as Principal Investigator at the Department of Computing Sciences, Bocconi University.

Interested in working on these problems? Bachelor’s and Master’s thesis projects building on DECODE-RNA are available, and I am always open to collaborations — get in touch.