Data Science Seminars: Bioinformatics focus
Events

Data Science Seminars: Bioinformatics focus

NOVEMBER 17, 2026

Featured image 1

Speaker: Carlos Moreno Silvestre

November 17th, 2026 | 2:30 pm
DEIB, PT2 Meeting Room (Bld. 20A)

For further information please contact:  Silvia Cascianelli |  silvia.cascianelli@polimi.it

Abstract

Tuesday, November 17th, 2026 at 2:30 pm a new appointment of Data Science Seminars: Bioinformatics focus will take place in DEIB PT2 Meeting Room (Building 20A), on line by Webex, organized by the Data Science for Bioinformatics group.

The seminar will be held by Carlos Moreno Silvestre on the following subject: "Methylome signatures of metastatic prostate cancer".

Metastatic castration-resistant prostate cancer is treated by suppressing androgen receptor (AR) signalling, yet some tumours escape by abandoning their luminal identity and adopting a neuroendocrine state that no longer depends on AR. Genomic alterations such as loss of RB1 and TP53 make this lineage switch possible, but the new state is thought to be maintained epigenetically, through the reprogramming of the enhancers that define cell identity. Here we asked whether lineage state is recorded in DNA methylation and where in the regulatory genome that record sits. We analysed whole-genome bisulphite sequencing of 99 metastatic biopsies from the West Coast Dream Team cohort, covering 26.1 million CpGs, with each biopsy labelled AR-active, intermediate or NEPC-like. Penalised regression models were trained for binary, three-class and ordinal tasks under stratified cross-validation, with every data-driven step confined to the training folds and significance assessed by rerunning the full analysis on shuffled labels. Methylation separated AR-active from NEPC-like biopsies with an out-of-fold ROC-AUC of 1.00 and reached a macro F1 of 0.79 across the three classes and a Spearman correlation of 0.76 with lineage order, values that no run with shuffled labels reached. The CpGs gaining methylation towards the neuroendocrine state were concentrated in adenocarcinoma regulatory elements and FOXA1 binding sites. Of the 103 CpGs of the ordinal model, 21 lay in adenocarcinoma elements, about ten times the expectation from background CpGs matched on regulatory status, methylation level, variance and distance to the transcription start site and none lay in neuroendocrine elements. Tumour purity differed too little between the lineages to account for more than about 5% of their methylation difference and neither somatic alterations, age nor biopsy site explained the signal. Together these results indicate that lineage plasticity is written into the methylation of luminal enhancers, consistent with the adenocarcinoma programme being silenced epigenetically as tumours move towards a neuroendocrine state. As the signal is carried by genomic regions rather than single CpGs, it suits targeted assays and once validated in an independent cohort it could allow the transition to be detected non-invasively from cell-free DNA.