Work

Identification of (ultra-)rare candidate functional promoter mutations in cancer using sequence-based deep learning models

Year: 2026

Type: article

Abstract: Background The identification of non-coding somatic cancer-driver mutations remains challenging due to difficulties in interpreting rare and ultra-rare variants. We hypothesized that sequence-based models can be used to systematically prioritize such mutations for their functional relevance. Methods Here, we developed a computational framework that leverages two sequence-based models to assess the functional impact of (ultra-)rare somatic single nucleotide variants (SNVs) in promoter regions. We first tested the validity of our approach by analysing SNVs derived from 3,825 whole-tumour genomes from patients for whom we also had matching RNA-seq data available. Results This very strict validation strategy yielded 57 genes that contain promoter mutations that our model consistently predicts to preferentially activate or repress gene expression in three cohorts of different patients. For nine genes we could confirm a nominally significant change using actual gene expression level measurements as well (6.32× higher than expected by chance, p -value = 1.16 × 10 − 5 ). These genes include well known cancer genes such as TERT and PMS2 , but also several new candidates for which no coding mutations have previously been implicated in cancer, including AIMP2 , SASS6 , ALKBH4 , FICD , TTK , YAE1 and ETF1 . We subsequently conducted a meta-analysis across 24,529 samples that revealed 619 genes, whose promoters are significantly enriched for candidate functional SNVs. These genes are significantly enriched for cancer-driver genes ( p -value = 5.94 × 10 − 3 ), cell cycle genes ( p -values ranging from 1.24 × 10 − 3 to 2.82 × 10 − 2 ) and are more evolutionary conserved than expected by chance ( p -value = 0.022). Conclusions These findings demonstrate the utility of sequence-based models for identifying candidate functional non-coding mutations and suggest that many non-coding somatic mutations contribute to cancer development.

Source: Genome Medicine

DOI: 10.1186/s13073-026-01821-9