Bradley E Bernstein | PubMed Health

PubMed2026

Loss of tumor-infiltrating lymphocytes and poor response to immunotherapy in IDH GOF mutant melanoma.

Recent innovations in melanoma treatment with immune checkpoint blockade (ICB) have improved overall outcomes for patients; however, over 50% of patients still develop resistance to treatment. These patients either have intrinsic resistance and never respond to therapy or develop acquired resistance months or years into treatment. The mechanisms underlying ICB resistance remain poorly understood. Our data show that patients with isocitrate dehydrogenase gain-of-function (IDH GOF) mutant melanoma have a worse response to anti-PD1 immunotherapy. IDH mutations have been found to be oncogenic and associated with differential methylation in multiple cancers but are not yet characterized in human melanoma. Here, we investigate the clinical, immune, and transcriptional phenotypes of IDH GOF melanomas through analyses of clinical response, single-cell RNA-seq, bulk RNA-seq, and DNA methylation data. Single-cell data analysis showed decreased immune infiltrate and activity in the IDH GOF tumors. Bulk sequencing data demonstrated the association among IDH mutation, immune exclusion, and disruptions in global DNA methylation. The melanoma-derived genomic data presented support previously described resistance mechanisms of IDH mutation in other cancer types and is the first demonstration to our knowledge of the role of IDH GOF in the human melanoma tumor microenvironment.

Humans↗

PubMed2025

A multi-modal transformer for cell type-agnostic regulatory predictions.

Sequence-based deep learning models have emerged as powerful tools for deciphering the cis-regulatory grammar of the human genome but cannot generalize to unobserved cellular contexts. Here, we present EpiBERT, a multi-modal transformer that learns generalizable representations of genomic sequence and cell type-specific chromatin accessibility through a masked accessibility-based pre-training objective. Following pre-training, EpiBERT can be fine-tuned for gene expression prediction, achieving accuracy comparable to the sequence-only Enformer model, while also being able to generalize to unobserved cell states. The learned representations are interpretable and useful for predicting chromatin accessibility quantitative trait loci (caQTLs), regulatory motifs, and enhancer-gene links. Our work represents a step toward improving the generalization of sequence-based deep neural networks in regulatory genomics.

Humans↗