Biology subjects

Fariselli, P.

Publications and source records attributed to Fariselli, P..

10 recordsLinked to original sources

bioRxiv2026

From {triangleup}{triangleup}G Prediction to Mutation Prioritization with Conformal Prediction

Selecting mutations for protein engineering and studies of protein function requires deciding which predicted stability changes warrant experimental follow-up. A point estimate alone does not show whether uncertainty leaves both stabilizing and destabilizing effects plausible. To address this limitation, we introduce ProbJanusDDG, which combines a protein-stability predictor with standard and adaptive conformal prediction intervals. Calibration assigns equal total weight to each protein, accounting for uneven numbers of measured mutations. For previously uncharacterized mutations, intervals lying entirely above or below zero provide a criterion for selecting candidate stabilizing or destabilizing substitutions, while intervals containing zero leave the direction unresolved. On the external low-homology S669L benchmark, adaptive intervals at the 50% nominal level assigned directional calls to 34.2% of mutations, with 94.8% observed sign accuracy. At the 70% level, 15.5% received calls, with 99.0% sign accuracy. At the 90% nominal level, standard and adaptive intervals achieved mutation-level coverage of 91.6% and 92.5%, whereas coverage averaged equally across proteins was approximately 87%. Adaptive intervals redistributed width across mutations, with greater widening among predictions with larger observed errors and nearly unchanged mean width. These results support uncertainty-aware screening of uncharacterized variants: prioritizing candidate stabilizing mutations for protein engineering and candidate destabilizing mutations for investigating stability-related effects on protein function. The method provides a selective criterion for experimental follow-up while distinguishing predictions with a resolved direction from those requiring further evidence.

bioinformatics↗

bioRxiv2026

A generative model for dimensionality reduction with millions of features and few samples

MotivationIn this paper, we demonstrate that it is feasible to train a deep generative model for dimensionality reduction with millions of features using few samples, which makes this type of generative model a more versatile alternative to standard methods for dimensionality reduction. Specifically, we hypothesize that for a decoder-only model, the number of training samples required is almost independent of the feature dimensionality in most network architectures. ResultsThrough an extensive set of experiments on synthetic non-linear data, we validate this hypothesis. We also train the model on a downsampled version of the 1000 Genomes Project (1KGP) dataset to further assess its behavior under controlled reductions in sample size. Furthermore, we train a deep generative decoder (DGD) on a curated dataset from the International Cancer Genome Consortium (ICGC), which contains 4.4 million features. It is trained on approximately 4,000 samples and tested on 1,000 samples. The resulting latent representation exhibits clear clustering, and when methods are reduced to the same number of dimensions, it outperforms PCA and VAE for tumor type classification. Additionally, the DGD is computationally efficient and can be trained on a 16GB GPU. Availability and implementationCode is available at https://github.com/cpancott/ReceptiveDGD. Contactcorrado.pancotti@helmholtz-munich.de; akrogh@di.ku.dk Supplementary informationSupplementary data are available with this preprint.

bioinformatics↗

bioRxiv2026

Generalized Chargaff symmetry in codon usage across the tree of life

Codon usage bias is a central record of mutation, selection, drift, and translational constraints, but it is usually treated separately from generalized Chargaff symmetry, the tendency for words and their reverse complements to occur at similar frequencies in long DNA sequences. Here we ask whether codon usage contains a measurable reverse-complement component, and whether departures from that component can be quantified. We first avoided imposing reverse-complement symmetry. Instead, we exhaustively evaluated all nontrivial factorized codon involutions. Across taxa, the reverse-complement transformation was the optimum, giving the highest median correlation between codon frequencies and transformed codon frequencies. Random and amino-acid-preserving reference models showed that the signal is not a generic property of codon profiles and is only partly explained by protein composition. Additional controls preserving amino-acid composition and matching GC3 in expectation showed that the observed reverse-complement correlation remains higher than expected from these constraints alone, and genus-level aggregation confirmed that the optimum is not driven by overrepresented genera. The symmetry breaks in a biologically ordered manner: the third, most degenerate codon position remains closest to the reverse-complement baseline, whereas the first departs most strongly and the second is intermediate and lineage dependent. Taxonomic comparisons reveal broad and fine-scale heterogeneity in codon-level symmetry preservation. Together, these results show that codon usage combines reverse-complement preservation with position-, lineage-, and function-dependent departures from the GCT-associated compositional baseline.

evolutionary biology↗

bioRxiv2026

The illusion of interpretability in biologically informed neural networks

Biologically informed neural networks (BINNs), also known as visible neural networks (VNNs), are widely adopted in omics because their architectures mirror known biological structures, such as gene-to-pathway relationships, and are therefore often assumed to be inherently interpretable. This assumption implies that learned gene-to-pathway weights and pathway node activations reflect meaningful biological mechanisms. Here, we show that this premise fails for a classical reason: nonidentifiability. Using a controlled teacher and student framework, we demonstrate that even under ideal conditions, including noiseless data, the correct model class, and identical sparse wiring, a BINN can perfectly recover the input-to-output mapping while failing to recover both gene-to-pathway weights and pathway activations. This failure persists across classification, regression, and survival tasks, and remains robust to variations in biological structure and network depth. Thus, the problem is not merely overparameterization or poor optimization: learning from outputs alone does not identify internal structure. Since biological mechanisms are not directly observed, recovering them from predictions alone is harder, not easier, than recovering neural network parameters, which are already known to be nonidentifiable. Critically, this failure reflects standard practice: widely used BINNs do not impose objective level constraints on gene-to-pathway weights or pathway activations, and therefore operate precisely in the regime modeled by our teacher-student framework. These results indicate that architectural transparency does not imply mechanistic interpretability. Without constraints that explicitly enforce identifiability, the apparent interpretability of BINNs reflects their design rather than what they actually learn.

bioinformatics↗

bioRxiv2025

Probing as a new technique to assess single-cell RNA-seq batch correction

Batch correction methods in single-cell RNA sequencing are essential for removing technical variation that can otherwise lead to misleading downstream analyses. The reliability of these methods is typically evaluated using unsupervised metrics. Here, we apply a Machine Learning (ML) technique called probing, formalized as the Batch Probing Score (BPS), to empirically demonstrate across six datasets that the most popular batch correctors fail to fully remove batch signal. In most cases, the batch of origin remains clearly identifiable after correction, even though standard evaluation metrics cannot detect it. We show that existing unsupervised metrics lack the sensitivity and specificity required to capture residual batch signal, whereas ML-based approaches can still detect it. This residual signal can similarly be picked up by downstream analysis tools, potentially leading to biased results. Because BPS is supervised, it directly quantifies batch signal strength by measuring how accurately the batch of origin can be predicted for each sample. It therefore provides an upper bound on the residual ML-actionable batch signal that could otherwise remain unnoticed. Our findings suggest that probing-based metrics should become a standard for assessing batch correction methods in single-cell RNA-seq and other areas of genomics.

bioinformatics↗

bioRxiv2024

The Specification Game: Rethinking the Evaluation of DrugResponse Prediction for Precision Oncology

Precision oncology plays a pivotal role in contemporary healthcare, aiming to optimize treatments for each patient based on their unique characteristics. This objective has spurred the emergence of various cancer cell line drug-response datasets, driven by the need to facilitate pre-clinical studies by exploring the impact of multi-omics data on drug response. Despite the proliferation of machine learning models for Drug Response Prediction (DRP), their validation remains critical to reliably assess their usefulness for drug discovery, precision oncology and their actual ability to generalize over the immense space of cancer cells and chemical compounds. This paper shows that the commonly used evaluation strategies for DRP methods learn solutions that optimize an unintended DRP score and fail to predict the proper drug-response activity ("specification gaming"). This problem hinders the advancement of the DRP field, and here we propose a new validation paradigm composed by three Aggregation Strategies (Global, Fixed-Drug, and Fixed-Cell Line) and three train-test Splitting Strategies to ensure a realistic assessment of the prediction performance. We also scrutinize the challenges associated with using IC50 as a prediction label, showing how its close correlation with the drug concentration ranges worsens the risk of misleading performance assessment. We thus propose also an alternative prediction label for DRP which is safer from this perspective.

bioinformatics↗

bioRxiv2023

MUSE-XAE: MUtational Signature Extraction with eXplainable AutoEncoder enhances tumour type classification

Mutational signatures are a critical component in deciphering the genetic alterations that underlie cancer development and have become a valuable resource for understanding the genomic changes that occur during tumorigenesis. In this paper, we present MUSE-XAE, a novel method for mutational signature extraction from cancer genomes using an explainable Auto-Encoder. Our approach employs a hybrid architecture consisting of a nonlinear encoder that can capture nonlinear interactions and a linear decoder, ensuring the interpretability of the active signatures in cancer genomes. We evaluated and compared MUSE-XAE with other available tools on synthetic and experimental cancer datasets and demonstrated that it achieves very accurate extraction capabilities while enhancing tumour-type classification. Our findings indicate that the use of Auto-Encoders is feasible and effective. This approach could facilitate further research in this area, with neural network-based models playing a critical role in advancing our understanding of cancer genomics

bioinformatics↗

bioRxiv2023

Recombulator-X: a fast and user-friendly tool for estimating X chromosome recombination rates in forensic genetics

Background and ObjectiveGenetic markers (especially short tandem repeats or STRs) located on the X chromosome are a valuable resource to solve complex kinship cases in forensic genetics in addition or alternatively to autosomal STRs. Groups of tightly linked markers are combined into haplotypes, thus increasing the discriminating power of tests. However, this approach requires precise knowledge of the recombination rates between adjacent markers. Recombination rates vary across the human genome and cannot be automatically derived from linkage physical maps. The International Society of Forensic Genetics recommends that recombination rate estimation on the X chromosome is performed from pedigree genetic data while taking into account the confounding effect of mutations. However, the only existing implementations that satisfy these requirements have several drawbacks: they were never publicly released, they are very slow and/or need cluster-level hardware and strong computational expertise to use. In order to address these key concerns, we developed Recombulator-X, a new open-source Python tool. MethodsThe most challenging issue, namely the running time, was addressed with dynamic programming techniques to greatly reduce the computational complexity of the algorithm, coupled with JIT compilation to further increase performance. We also extended the statistical framework from STR to any polymorphic marker. ResultsCompared to the previous methods, Recombulator-X reduces the estimation times from weeks or months to less than one hour for typical datasets. Moreover, the estimation process, including preprocessing, has been streamlined and packaged into a simple command-line tool that can be run on a normal PC. Where previous approaches were limited to small panels of STR markers (up to 15), our tool can handle greater numbers (up to 100) of mixed STR and non-STR markers. ConclusionsIn the genetic forensic community, state-of-the-art estimation methods for X chromosome recombination rates have seen limited usage due to the technical hurdles posed by previous implementations. Recombulator-X makes the process much simpler, faster and accessible to researchers without a computational background, hopefully spurring increased adoption of best practices. Moreover, it extends the estimation framework to larger panels of genetic markers (not only STRs), allowing analyses of sequencing-based data.

bioinformatics↗

bioRxiv2021

Context dependency of nucleotide probabilities and variants in human DNA

BackgroundGenomic DNA has been shaped by mutational processes through evolution. The cellular machinery for error correction and repair has left its marks in the nucleotide composition along with structural and functional constraints. Therefore, the probability of observing a base in a certain position in the human genome is highly context-dependent. ResultsHere we develop context-dependent nucleotide models. We first investigate models of nucleotides conditioned on sequence context. We develop a bidirectional Markov model that use an average of the probability from a Markov model applied to both strands of the sequence and thus depends on up to 14 bases to each side of the nucleotide. We show how the genome predictability varies across different types of genomic regions. Surprisingly, this model can predict a base from its context with an average of more than 50% accuracy. For somatic variants we show a tendency towards higher probability for the variant base than for the reference base. Inspired by DNA substitution models, we develop a model of mutability that estimates a mutation matrix (called the alpha matrix) on top of the nucleotide distribution. The alpha matrix can be estimated from a much smaller context than the nucleotide model, but the final model will still depend on the full context of the nucleotide model. With the bidirectional Markov model of order 14 and an alpha matrix dependent on just one base to each side, we obtain a model that compares well with a model of mutability that estimates mutation probabilities directly conditioned on three nucleotides to each side. For somatic variants in particular, our model fits better than the simpler model. Interestingly, the model is not very sensitive to the size of the context for the alpha matrix. ConclusionsOur study found strong context dependencies of nucleotides in the human genome. The best model uses a context of 14 nucleotides to each side. Based on these models, a substitution model was constructed that separates into the context model and a matrix dependent on a small context. The model fit somatic variants particularly well.

bioinformatics↗

bioRxiv2019

DNA energy constraints shape biological evolutionary trajectories

Most living systems rely on double-stranded DNA (dsDNA) to store their genetic information and perpetrate themselves. Thus, the biological information contained within a dsDNA molecule, in terms of a linear sequence of nucleotides, has been considered the main target of the evolution. However, in this information-centred perspective, certain DNA sequence symmetries are difficult to explain. Here we show that these patterns can emerge from the physical peculiarities of the dsDNA molecule itself and the maximum entropy principle alone, rather than from biological or environmental evolutionary pressure. Our predictions are valid for both prokaryotes and eukaryotes, and also inform the interpretation of observed codon biases and context-dependent mutation patterns in human populations. Our results suggest that the double helix energy constraints and, more generally, the physical properties of the dsDNA are the hard drivers of the overall DNA sequence architecture, whereas the biological selective processes act as soft drivers, which only under extraordinary circumstances overtake the overall entropy content of the genome.

molecular biology↗