Piero Fariselli

The paper introduces ProbJanusDDG to prioritize mutations for experimental follow-up by using conformal prediction intervals to assess uncertainty in predicted stability changes.

This paper introduces a deep generative decoder model for dimensionality reduction with millions of features and few samples, showing it outperforms PCA and VAE in tumor classification.

The paper quantifies reverse-complement symmetry in codon usage across taxa and shows it is not fully explained by compositional constraints.

This paper shows that biologically informed neural networks (BINNs) cannot reliably recover internal biological structures from outputs alone due to nonidentifiability.

This paper shows that popular scRNA-seq batch correction methods fail to fully remove batch signal, which can affect downstream analyses, and introduces a supervised metric, BPS, to better assess correction effectiveness.

This paper identifies flaws in current Drug Response Prediction evaluation methods and proposes new strategies to improve their reliability and generalization.

MUSE-XAE extracts mutational signatures from cancer genomes using an explainable Auto-Encoder, achieving accurate extraction and improving tumour-type classification.

Recombulator-X estimates X chromosome recombination rates from pedigree data, reducing computation time from weeks to under an hour and supporting larger marker panels.

The paper develops context-dependent nucleotide models for the human genome, showing they can predict bases with over 50% accuracy and fit somatic variants better than simpler models.

This paper shows that DNA sequence patterns arise from physical constraints of dsDNA and maximum entropy, not evolutionary pressure.