Biology subjects

Fortino, V.

Publications and source records attributed to Fortino, V..

6 recordsLinked to original sources

bioRxiv2026

Multi-objective domain adaptation for pan-cancer transfer of pharmacogenomic drug response to patient tumours

Large-scale pharmacogenomic screens provide rich drug-sensitivity measurements in cancer cell lines, but transferring these measurements to patient tumours remains limited by biological heterogeneity and cross-domain distribution shift. Existing machine learning approaches, including adversarial autoencoder and domain adaptation, address this gap by learning shared latent representations between preclinical models and patient samples. However, alignment without explicit biological constraints can produce unreliable representations when cancer lineage structure and clinically relevant patient heterogeneity are not preserved. Here, we introduce MODAE+, a pan-cancer multi-objective deep learning domain adaptation framework that jointly balances domain alignment, deconfounding, drug-response prediction, survival prediction, and cancer-lineage classification across 9,696 TCGA tumours, 1,293 CCLE cancer cell lines, and CTRP sensitivity profiles for 544 compounds in 738 matched cell lines. This multi-objective design improved cross-domain alignment while preserving cancer-lineage organization and patient heterogeneity. For drugs shared between the pharmacogenomic screen and TCGA treatment records, MODAE+ predicted sensitivity was significantly associated with overall survival among patients who received each drug in 46.3% of evaluated cancer-drug pairs by log-rank testing and 41.5% by multivariable Cox analysis. Biological validation across four curated drug-gene resources identified significant expression differences exceeding 1.5-fold for 34.9-64.4% of drugs in pan-cancer analyses, based on the lowest-p-value annotated gene per drug. Drug-class-specific patterns involved intended molecular targets for targeted therapies and predominantly transport, metabolism, detoxification and resistance-related genes for classical chemotherapies. As an additional validation, we derived pretreatment responder-nonresponder gene-expression signatures from 13 external breast cancer datasets and evaluated their concordance with MODAE+ predictions across 69 signature-drug comparisons in TCGA patients treated with paclitaxel or doxorubicin. Consensus signature scores showed Spearman correlations of 0.791 with predicted paclitaxel sensitivity and 0.724 with predicted doxorubicin sensitivity, averaging 0.758 across the two drugs. Together, these complementary analyses provide supporting evidence for the clinical relevance and biological consistency of MODAE+ predictions, suggesting their potential to identify drug-sensitive tumour subgroups and generate hypotheses about mechanisms of drug resistance.

bioinformatics↗

bioRxiv2024

Network-based estimation of therapeutic efficacy and adverse reaction potential for prioritisation of anti-cancer drug combinations

Drug combinations, although a key therapeutic agent against cancer, are yet to reach their full applicability potential due to the challenges involved in the identification of effective and safe drug pairs. In vitro or in vivo screening would have been the optimal approach if combinatorial explosion was not an issue. In silico methods, on the other hand, can enable rapid screening of drug pairs to prioritise for experimental validation. Here we present a novel network medicine approach that systematically models the proximity of drug targets to disease-associated genes and adverse effect- associated genes, through the combination of network propagation algorithm and gene set enrichment analysis. The proposed approach is applied in the context of identifying effective drug combinations for cancer treatment starting from a training set of drug combinations curated from DrugComb and DrugBank databases. We observed that effective drug combinations usually enrich disease-related gene sets while adverse drug combinations enrich adverse-effect gene sets. We use this observation to systematically train classifiers distinguishing drug combinations with higher therapeutic effects and no known adverse reaction from combinations with lower therapeutic effects and potential adverse reactions in six cancer types. The approach is tested and validated using drug combinations curated from in vitro screening data and clinical reports. Trained classification models are also used to identify novel potential anti-cancer drug combinations for experimental validation. We believe our framework would be a key addition to the anti-cancer drug combination identification pipeline by enabling rapid yet robust estimation of therapeutic efficacy or adverse reaction potential.

bioinformatics↗

bioRxiv2024

Multi-task deep latent spaces for cancer survival and treatment response prediction

MotivationCancer is a very heterogeneous disease that can be difficult to treat without addressing the specific mechanisms driving tumour progression in a given patient. High-throughput screening and sequencing data from cancer cell-lines has driven many developments in drug development, however, there are important aspects crucial to precision medicine that are often overlooked, namely the inherent differences between tumours in patients and the cell-lines used to model them in vitro. Recent developments in transfer learning methods for patient and cell-line data have shown progress in translating results from cell-lines to individual patients in silico. However, transfer learning can be forceful and there is a risk that clinically relevant patterns in the omics profiles of patients are lost in the process. ResultsWe present MODAE, a novel deep learning algorithm to integrate omics profiles from cell-lines and patients for the purposes of exploring precision medicine opportunities. MODAE implements patient survival prediction as an additional task in a drug-sensitivity transfer learning schema and aims to balance autoencoding, domain adaptation, drugsensitivity prediction, and survival prediction objectives in order to better preserve the heterogeneity between patients that is relevant to survival. While burdened with these additional tasks, MODAE performed on par with baseline survival models, but struggled in the drug-sensitivity prediction task. Nevertheless, these preliminary results were promising and show that MODAE provides a novel AI-based method for prioritizing drug treatments for high-risk patients. Availabilityhttps://github.com/UEFBiomedicalInformaticsLab/MODAE

bioinformatics↗

bioRxiv2022

Comparison of missing data handling methods for variant pathogenicity predictors

1BackgroundModern clinical genetic tests utilize next-generation sequencing (NGS) approaches to comprehensively analyze genetic variants from patients. Out of these millions of variants, clinically relevant variants that match the patients phenotype need to be identified accurately within a rapid timeframe that facilitates clinical action. As manual evaluation of variants is not a feasible option for meeting the speed and volume requirements of clinical genetic testing, automated solutions are needed. Various machine learning (ML), artificial intelligence (AI), and in silico variant pathogenicity predictors have been developed to solve this challenge. These solutions rely on the comprehensiveness of the available data and struggle with the sparse nature of genetic variant data. Therefore, careful treatment of missing data is necessary, and the selected methods may have a huge impact on the accuracy, reliability, speed and associated computational costs. ResultsWe present an open-source framework called AMISS that can be used to evaluate performance of different methods for handling missing genetic variant data in the context of variant pathogenicity prediction. Using AMISS, we evaluated 14 methods for handling missing values. The performance of these methods varied substantially in terms of precision, computational costs, and other attributes. Overall, simpler imputation methods and specifically mean imputation performed best. ConclusionsSelection of the missing data handling method is crucial for AI/ML-based classification of genetic variants. We show that utilizing sophisticated imputation methods is not worth the cost when used in the context of genetic variant pathogenicity classification.

bioinformatics↗

bioRxiv2021

The integration of large-scale public data and network analysis uncovers molecular characteristics of psoriasis

In recent years, a growing interest in the characterization of the molecular basis of psoriasis has been observed. However, despite the availability of a large amount of molecular data, many pathogenic mechanisms of psoriasis are still poorly understood. In this study, we performed an integrated analysis of 23 public transcriptomic datasets encompassing both lesional and uninvolved skin samples from psoriasis patients. We defined comprehensive gene co-expression network models of psoriatic lesions and uninvolved skin. Moreover, we collected, curated and exploited a wide range of functional information from multiple public sources in order to systematically annotate the inferred networks. The integrated transcriptomics analysis of public datasets shed light on a number of genes which are frequently deregulated in the psoriatic lesion compared with the unaffected skin in a large number of studies. In particular, CRABP2, LCN2, S100A12 and PDZK1IP1 were found to be deregulated in all of the datasets analyzed. Furthermore, the analysis of co-expression networks highlights genes showing aberrant patterns of connectivity in the lesional network as compared to the network inferred from unaffected skin samples. For instance, we identified co-expression patterns of SERPINB4, KYNU and S100A12 as being the most affected by the disease. Network analysis allowed us to identify YPEL1 and HUS1 as plausible, previously unknown, actors in the expression of the psoriasis phenotype. In addition, by exploiting topological properties of the network models, we highlighted a set of 250 non-deregulated genes, 223 of which have never been associated with the disease before, including CACNA1A, HADH, ATP5MC1 and CBARP among others. Finally, we characterized specific communities of co-expressed genes sustaining relevant molecular functions and specific immune cell types expression signatures playing a role in the psoriasis lesion. Overall, integrating experimental driven results with curated functional information from public repositories represents an efficient approach to empower knowledge generation about psoriasis and may be applicable to other complex diseases.

systems biology↗

bioRxiv2018

eUTOPIA: Solution for omics data preprocessing and analysis

Application of microarrays in omics technologies enables quantification of many biomolecules simultaneously. It is widely applied to observe the positive or negative effect on biomolecule activity in perturbed versus the steady state by quantitative comparison. Community resources, such as Bioconductor and CRAN, host tools based on R that have become standard for high-throughput analytics. However, there is a need for intuitive and easy-to-use platform to process omics data, visualize, and interpret results, which is computational skill neutral. We propose an integrated software solution, eUTOPIA, that implements a set of essential processing steps as a guided workflow presented to the user as an R Shiny application. eUTOPIA allows researchers to perform preprocessing and analysis of microarray data via a simple and intuitive graphical interface while using state of the art methods. eUTOPIA is free for academic use and can be obtained from the GitHub repository https://github.com/Greco-Lab/eUTOPIA.

bioinformatics↗