Biology subjects

Ballester, P.

Publications and source records attributed to Ballester, P..

16 recordsLinked to original sources

bioRxiv2026

Systematic Evaluation of Transfer Learning Strategies for Clinical Chemotherapy Response Prediction

Accurately predicting chemotherapy response remains a major challenge in precision oncology. Although machine-learning models based on tumour omics data have shown promise, the majority of existing studies are trained and evaluated on pre-clinical cell-line datasets, leaving their clinical applicability insufficiently characterised. In this study, we systematically evaluate a range of transfer-learning strategies for chemotherapy response prediction under realistic clinical constraints using patient data from The Cancer Genome Atlas (TCGA). Rather than proposing a new predictive model, we focus on assessing the effectiveness and limitations of commonly used approaches for transferring pre-clinical knowledge to clinical settings. These include cell-line-validated biomarkers, biologically informed feature representations, direct application of pre-clinical deep-learning models, model fine-tuning, and hybrid strategies that integrate pre-clinical predictions with clinical data. All methods are evaluated within a unified framework using consistent cohort construction, shared performance metrics, and bias-controlled validation procedures. Across multiple drugs and molecular data types, we find that most transfer strategies--including biomarker-based feature selection and direct pre-clinical model transfer--fail to produce robust or consistent improvements in clinical prediction performance. In contrast, conservative approaches based on fine-tuning pre-clinical models or incorporating pre-clinical predictions as features in clinical models yield more stable and reproducible gains. Further improvements are observed when basic pre-treatment clinical variables are integrated. Together, our results demonstrate the practical boundaries of pre-clinical to clinical transfer for drug response prediction and highlight hybrid and fine-tuning strategies as more reliable baselines for future translational modelling efforts.

bioinformatics↗

bioRxiv2025

A Pan-Cancer, Pan-Treatment (PCPT) Model For Predicting Drug Responses From Patient-Derived Xenografts With Application To Cancer Types With Sparse Training Data

Predicting personalized treatment-specific responses in cancer patients requires not only robust experimental models such as PDX, which generate accurate training data on different treatment responses for the same model, but also computational models that can translate this knowledge into decisions for real patients in a timely manner. The translatability of PDX data into patient-specific outcomes is limited by the challenges in the data size requirements for training machine learning models. Previously, ML models have been developed for the two most abundant cancer types, viz. BRCA and CRC, but the same could not be scaled up to other cancer types with a smaller number of PDX samples. Here, we provide an ML framework to train a single pan-cancer, pan-treatment model for predicting treatment outcomes. We show that such models give promising results for all cancer types considered and reproduce the accuracy levels of individually trained cancer types. In the proposed model, all PDX genomic profiles from all cancer types are used as the training data, and instead of partitioning them into cancer types for each model, the cancer type and treatment name are appended as the input features of the training model. Using genomic-only and treatment-only embeddings and combining them with PCA-based dimensionality reduction, our models show promising results and provide a framework for further improvements and real-time use for best treatment selections for cancer patients. Data and Code AvailabilityAll raw and processed data, along with analysis scripts, are available at: https://github.com/Sciwhylab/pcpt

bioinformatics↗

bioRxiv2025

Machine Learning-Based Drug Response Prediction Identifies Novel Therapeutic Candidates for Colorectal Cancer Cell Line KM-12

One major hurdle in the discovery of new drugs is the limited capacity of traditional screening methods to efficiently explore the immense space of drug candidates. To bypass this difficulty, computational approaches such as virtual screening (VS) have been developed supported by Machine Learning (ML) models to predict the activities of drug-like molecules on a given target. We developed a ML model for VS of a library of 25 million highly-diverse synthesis-on-demand molecules against the KM-12 colorectal cancer (CRC) cell line. By combining different biological readouts, we discovered compounds with a novel chemical scaffold, not yet described in the context of CRC, with either a prominent cytotoxic or cytostatic effect. This finding underpins the strength of our ML-guided VS protocol in discovering promising CRC drug leads triggering altered biological responses.

cancer biology↗

bioRxiv2024

Predicting temozolomide response in low-grade glioma patients with large-scale machine learning

BackgroundTemozolomide is the primary chemotherapeutic agent and first-line treatment for low-grade glioma (LGG). Although LGGs are generally less aggressive than high-grade gliomas (HGGs), they can eventually progress into HGGs, making it crucial to maximise the efficacy of initial LGG treatment. MethodsIn this study, we analysed data from 109 LGG patients in The Cancer Genome Atlas (TCGA) to assess the predictive performance of 12 machine learning (ML) classification algorithms in forecasting temozolomide response, utilising six types of omics data. Cross-validation and bootstrapping bias correction were employed to compare the performance of these models with that of a conventional biomarker-based model using MGMT promoter methylation status. ResultsAmong the models, the miRNA-based approach using the XGBoost algorithm showed the most promising predictive performance, with a Matthews Correlation Coefficient (MCC) of 0.447, outperforming the auto-ML method JADBio (MCC = 0.250) and the MGMT biomarker-based model. Converting the task into a regression framework by encoding cancer response as a continuous variable generally weakened predictive performance, with best model based on methylation profile (MCC = 0.344). Additionally, ML models focusing on biomarker-relevant features demonstrated improved prediction accuracy, though not to the extent of omics-based models, with the support vector machine (SVM) model achieving an MCC of 0.331. Incorporating clinical variables, such as patient age and Karnofsky score, further enhanced predictive power, with the LR-OMC model achieving the highest MCC (0.483). Feature importance analysis identified six significant miRNA factors, including three tumour-related miRNAs (miR-335, let-7f, and miR-7-2) and three potential biomarkers (miR-204, miR-6513, and miR-376). DiscussionOverall, this study systematically demonstrates the potential of large-scale analyses combining machine learning and omics data to predict temozolomide response, offering superior predictive accuracy compared to standard MGMT biomarkers.

bioinformatics↗

bioRxiv2024

DNA methylation is a key determinant of response to targeted and immune checkpoint therapies in metastatic renal cell carcinoma

The response to targeted therapies and immune checkpoint inhibitors for patients suffering from metastatic clear cell renal cell carcinoma (ccRCC) is heterogeneous and currently not predictable in clinic. In this work, a comprehensive integrated study of 700 ccRCCs profiled by DNA methylation and RNA sequencing showed that the hyper-methylated tumors exhibited a worse prognosis, a higher fraction of cycling tumor cells and a lower activity of homeobox transcription factors. To translate the use of DNA methylation information into a clinical setting, we developed a simple model accurately predicting the ccRCC methylation subtypes (AUC-ROCs of 0.91) from two gene expression ratios (IGF2BP3/PCCA, TNNT1/TMEM88). In addition, these methylation subtypes were significantly associated with the therapeutic outcome of patients to anti-PD-1, mTOR inhibitor or tyrosine kinase inhibitor therapies. Overall, our framework for predicting the ccRCC DNA methylation subtypes from targeted gene expression data is easy to translate in clinic and contributes to better personalization of ccRCC therapies.

bioinformatics↗

bioRxiv2024

Graph neural networks best guide phenotypic virtual screening on cancer cell lines

Artificial intelligence is increasingly driving early drug design, offering novel approaches to virtual screening. Phenotypic virtual screening (PVS) aims to predict how cancer cell lines respond to different compounds by focusing on observable characteristics rather than specific molecular targets. Some studies have suggested that deep learning may not be the best approach for PVS. However, these studies are limited by the small number of tested molecules as well as not employing suitable performance metrics and dissimilar-molecules splits better mimicking the challenging chemical diversity of real-world screening libraries. Here we prepared 60 datasets, each containing approximately 30,000 to 50000 molecules tested for their growth inhibitory activities on one of the NCI-60 cancer cell lines. We evaluated the performance of five machine learning algorithms for PVS on these 60 problem instances. To provide a comprehensive evaluation, we employed two model validation types: the random split and the dissimilar-molecules split. The models were primarily evaluated using hit rate, a more suitable metric in VS contexts. The results show that all models are more challenged by test molecules that are substantially different from those in the training data. In both validation types, the D-MPNN algorithm, a graph-based deep neural network, was found to be the most suitable for building predictive models for this PVS problem.

bioinformatics↗

bioRxiv2024

Conformal prediction of molecule-induced cancer cell growth inhibition challenged by strong distribution shifts

The drug discovery process often employs phenotypic and target-based virtual screening to identify potential drug candidates. Despite the longstanding dominance of target-based approaches, phenotypic virtual screening is undergoing a resurgence due to its potential being now better understood. In the context of cancer cell lines, a well-established experimental system for phenotypic screens, molecules are tested to identify their whole-cell activity, as summarized by their half-maximal inhibitory concentrations. Machine learning has emerged as a potent tool for computationally guiding such screens, yet important research gaps persist, including generalization and uncertainty quantification. To address this, we leverage a clustering-based validation approach, called Leave Dissimilar Molecules Out (LDMO). This strategy enables a more rigorous assessment of model generalization to structurally novel compounds. This study focuses on applying Conformal Prediction (CP), a model-agnostic framework, to predict the activities of novel molecules on specific cancer cell lines. A total of 4320 independent models were evaluated across 60 cell lines, 5 CP variants, 2 set features, and training-test splits, providing strong and consistent results. From this comprehensive evaluation, we concluded that, regardless of the cell line or model, novel molecules with smaller CP-calculated confidence intervals tend to have smaller predicted errors once measured activities are revealed. It was also possible to anticipate the activities of dissimilar test molecules across 50 or more cell lines. These outcomes demonstrate the robust efficacy that LDMO-based models can achieve in realistic and challenging scenarios, thereby providing valuable insights for enhancing decision-making processes in drug discovery.

bioinformatics↗

bioRxiv2024

Comprehensive machine learning boosts structure-based virtual screening for PARP1 inhibitors

Poly ADP-ribose polymerase 1 (PARP1) is an attractive therapeutic target for cancer treatment. Machine-learning scoring functions constitute a promising approach to discovering novel PARP1 inhibitors. Cutting-edge PARP1-specific machine-learning scoring functions were investigated using semi-synthetic training data from docking activity-labelled molecules: known PARP1 inhibitors, hard-to-discriminate decoys property-matched to them with generative graph neural networks and confirmed inactives. We further made test sets harder by including only molecules dissimilar to those in the training set. Comprehensive analysis of these datasets using five supervised learning algorithms, and protein-ligand fingerprints extracted from docking poses and ligand only features revealed two highly predictive scoring functions. The PARP1-specific support vector machine-based regressor, when employing PLEC fingerprints, achieved a high Normalized Enrichment Factor at the top 1% on the hardest test set (NEF1% = 0.588, median of 10 repetitions), and was more predictive than any other investigated scoring function, especially the classical scoring function employed as baseline. Scientific ContributionWe present the first PARP1-specific machine-learning scoring functions for structure-based virtual screening. A particularly rigorous evaluation, including test sets with novel molecules and a much higher proportion of challenging property-matched decoys, reveals the most predictive scoring function for this important therapeutic target. Typically, narrow machine learning analyses would have likely missed this promising PARP1-specific scoring function, which is now released with this paper so that others can use it for prospective virtual screening. Key PointsO_LIA new scoring tool based on machine-learning was developed to predict PARP1 inhibitors for potential cancer treatment. C_LIO_LIThe majority of PARP1-specific machine-learning models performed better than generic and classical scoring functions. C_LIO_LIAugmenting the training set with ligand-only Morgan fingerprint features generally resulted in better performing models, but not for the best models where no further improvement was observed. C_LIO_LIEmploying protein-ligand-extracted fingerprints as molecular descriptors led to the best-performing and most-efficient model for predicting PARP1 inhibitors. C_LIO_LIDeep learning performed poorly on this target in comparison with the simpler ML models. C_LI

bioinformatics↗

bioRxiv2020

An NF-κB/IRF1 axis programs cDC1s to drive anti-tumor immunity

Conventional type 1 dendritic cells (cDC1s) are critical for anti-tumor immunity. They acquire antigens from dying tumor cells and cross-present them to CD8+ T cells, promoting the expansion of tumor-specific cytotoxic T cells. However, the signaling pathways that govern the anti-tumor functions of cDC1s are poorly understood. We mapped the molecular pathways regulating intra-tumoral cDC1 maturation using single cell RNA sequencing. We identified NF-{kappa}B and IFN pathways as being highly enriched in a subset of functionally mature cDC1s. The specific targeting of NF-{kappa}B or IFN pathways in cDC1s prevented the recruitment and activation of CD8+ T cells and the control of tumor growth. We identified an NF-{kappa}B-dependent IFN{gamma}-regulated gene network in cDC1s, including cytokines and chemokines specialized in the recruitment and activation of cytotoxic T cells. We used single cell transcriptomes to map the trajectory of intra-tumoral cDC1 maturation which revealed the dynamic reprogramming of tumor-infiltrating cDC1s by NF-{kappa}B and IFN signaling pathways. This maturation process was perturbed by specific inactivation of either NF-{kappa}B or IRF1 in cDC1s, resulting in impaired expression of IFN-{gamma}-responsive genes and consequently a failure to efficiently recruit and activate anti-tumoral CD8+ T cells. Finally, we demonstrate the relevance of these findings to cancer patients, showing that activation of the NF-{kappa}B/IRF1 axis in association with cDC1s is linked with improved clinical outcome. The NF-{kappa}B/IRF1 axis in cDC1s may therefore represent an important focal point for the development new diagnostic and therapeutic approaches to improve cancer immunotherapy. One Sentence SummaryNF-{kappa}B and IRF1 coordinate intra-tumoral cDC1 maturation and control of immunogenic tumor growth.

immunology↗

bioRxiv2020

Characterizing the relationship between the chemical structures of drugs and their activities on primary cultures of pediatric solid tumors

Better drugs are required to manage pediatric cancers. A high-throughput screen of drugs in primary cultures derived from orthotopic patient-derived xenografts (O-PDX) of pediatric solid tumours has been recently published. Here we analyzed these data sets to find out whether it is possible to leverage them for identifying new drug leads in a phenotypic manner. We found that drugs bearing a higher number of heterocyclic rings, two carbon-hetero bonds and halogens are associated to submicromolar potency in alveolar rhabdomyosarcoma and osteosarcoma O-PDXs. Furthermore, Murcko scaffolds 1-cyclopentyl-octahydro-1H-indene and tetradecahydroanthracene can be utilized as starting scaffolds to selectively optimize potency against osteosarcoma since drugs bearing this scaffold displayed superior O-PDX culture potency. Lastly, we have generated QSAR (Quantitative Structure-Activity Relationship) models able to predict the potency of drugs on each O-PDX tumor. To permit their use to guide drug repositioning on these 30 O-PDX cell cultures, we are providing a user-friendly web server implementing these QSAR models at https://rnewbie.shinyapps.io/Shobek-master

pharmacology and toxicology↗

bioRxiv2020

In silico molecular target prediction unveils mebendazole as a potent MAPK14 inhibitor

The concept of polypharmacology involves the interaction of drug molecules with multiple molecular targets. It provides a unique opportunity for the repurposing of already-approved drugs to target key factors involved in human diseases. Herein, we used an in silico target prediction algorithm to investigate the mechanism of action of mebendazole, an anti-helminthic drug, currently repurposed in the treatment of brain tumors. First, we confirmed that mebendazole decreased the viability of glioblastoma cells in vitro. Our in silico approach unveiled 21 putative molecular targets for mebendazole, including 12 proteins significantly up-regulated at the gene level in glioblastoma as compared to normal brain tissue. Validation experiments were performed on three major kinases involved in cancer biology: ABL1, MAPK1/ERK2 and MAPK14/p38α. Mebendazole could inhibit the activity of these kinases in vitro in a dose-dependent manner, with a high potency against MAPK14. Its direct binding to MAPK14 was further validated in vitro and inhibition of MAPK14 kinase activity was confirmed in live glioblastoma cells. Consistent with biophysical data, molecular modeling suggested that mebendazole was able to bind to the catalytic site of MAPK14. Finally, gene silencing demonstrated that MAPK14 is involved in glioblastoma tumor spheroid growth and response to mebendazole treatment. This study thus highlighted the role of MAPK14 in the anticancer mechanism of action of mebendazole and provides further rationale for the pharmacological targeting of MAPK14 in brain tumors. It also opens new avenues for the development of novel MAPK14/p38α inhibitors to treat human diseases.Significance Statement This study provides a framework to investigate drug polypharmacology by rapidly identifying novel molecular targets of already-approved drugs. It unveils a new mechanism involved in the anticancer activity of anti-helminthic drug, mebendazole, which is currently being repurposed for the treatment of brain tumors. By helping to decipher the mechanism(s) of action of repurposed drugs in their new indications, this approach could contribute to the development of safer and more effective therapeutic strategies in oncology and beyond.Competing Interest StatementThe authors have declared no competing interest.View Full Text

pharmacology and toxicology↗

bioRxiv2020

The impact of compound library size on the performance of scoring functions for structure-based virtual screening

Larger training datasets have been shown to improve the accuracy of Machine Learning (ML)-based Scoring functions (SFs) for Structure-Based Virtual Screening (SBVS). In addition, massive test sets for SBVS, known as ultra-large compound libraries, have been demonstrated to enable the fast discovery of selective drug leads with at least nanomolar potency. This proof-of-concept was carried out on two targets using a single docking tool along with its SF. It is thus unclear whether this high level of performance would generalise to other targets, docking tools and SFs. We found that screening a larger compound library results in more potent actives being identified in all six additional targets using a different docking tool along with its classical SF. Furthermore, we established that a way to improve the potency of the retrieved molecules further is to rank them with more accurate ML-based SFs (we found this to be true in four of the six targets, the difference was not significant in the remaining two targets). A three-fold increase in average hit rate across targets was also achieved by the ML-based SFs. Lastly, we observed that classical and ML-based SFs often find different actives, which supports using both types of SFs on those targets. Contactpedro.ballester@inserm.fr Supplementary informationan online-only supplementary results file is enclosed. Biographical NoteL. Fresnais carried out a master research project directly supervised by P.J Ballester and he will soon be starting a PhD. P.J Ballester has been working on virtual screening for over 15 years now. He is group leader and research scientist at cancer research centre of INSERM, the French National Institute of Health & Medical Research.

bioinformatics↗

bioRxiv2018

Predicting synergism of cancer drug combinations using NCI-ALMANAC data

BackgroundDrug combinations are of great interest for cancer treatment. Unfortunately, the discovery of synergistic combinations by purely experimental means is only feasible on small sets of drugs. In silico modeling methods can substantially widen this search by providing tools able to predict which of all possible combinations in a large compound library are synergistic. Here we investigate to which extent drug combination synergy can be predicted by exploiting the largest available dataset to date (NCI-ALMANAC, with over 290,000 synergy determinations).\n\nMethodsEach cell line is modeled using primarily two machine learning techniques, Random Forest (RF) and Extreme Gradient Boosting (XGBoost), on the datasets provided by NCI-ALMANAC. This large-scale predictive modeling study comprises more than 5000 pair-wise drug combinations, 60 cell lines, 4 types of models and 5 types of chemical features. The application of a powerful, yet uncommonly used, RF-specific technique for reliability prediction is also investigated.\n\nResultsThe evaluation of these models shows that it is possible to predict the synergy of unseen drug combinations with high accuracy (Pearson correlations between 0.43 and 0.86 depending on the considered cell line, with XGBoost providing slightly better predictions than RF). We have also found that restricting to the most reliable synergy predictions results in at least two-fold error decrease with respect to employing the best learning algorithm without any reliability estimation. Alkylating agents, tyrosine kinase inhibitors and topoisomerase inhibitors are the drugs whose synergy with other partner drugs are better predicted by the models.\n\nConclusionsDespite its leading size, NCI-ALMANAC comprises an extremely small part of all conceivable combinations. Given their accuracy and reliability estimation, the developed models should drastically reduce the number of required in vitro tests by predicting in silico which of the considered combinations are likely to be synergistic.

bioinformatics↗

bioRxiv2018

Machine learning models to predict in vivo drug response via optimal dimensionality reduction of tumour molecular profiles

Inter-tumour heterogeneity is one of cancers most fundamental features. Patient stratification based on drug response prediction is hence needed for effective anti-cancer therapy. However, lessons from the past indicate that single-gene markers of response are rare and/or often fail to achieve a significant impact in clinic. In this context, Machine Learning (ML) is emerging as a particularly promising complementary approach to precision oncology. Here we leverage comprehensive Patient-Derived Xenograft (PDX) pharmacogenomic data sets with dimensionality-reducing ML algorithms with this purpose. Results show that combining multiple gene alterations via ML leads to better discrimination between sensitive and resistant PDXs in 19 of the 26 analysed cases. Highly predictive ML models employing concise gene lists were found for three cases: Paclitaxel (breast cancer), Binimetinib (breast cancer) and Cetuximab (colorectal cancer). Interestingly, each of these ML models identify some responsive PDXs not harbouring the best actionable mutation for that case (such PDXs were missed by those single-gene markers). Moreover, ML multi-gene predictors generally retrieve a much higher proportion of treatment-sensitive PDXs than the corresponding single-gene marker. As PDXs often recapitulate clinical outcomes, these results suggest that many more patients could benefit from precision oncology if multiple ML algorithms were applied to existing clinical pharmacogenomics data, especially those algorithms generating classifiers combining data-selected gene alterations.

bioinformatics↗

bioRxiv2017

Cancer Cell Line Profiler (CCLP): a webserver for the prediction of compound activity across the NCI60 panel

SummaryCCLP (Cancer Cell Line Profiler) is a webserver for the prediction of compound activity across the NCI60 panel. CCLP uses a multi-task Random Forest model trained on 941,831 data-points that integrates structural information from 17,142 compounds and multi-omics data sets from 59 cancer cell lines. In addition, CCLP also implements conformal prediction to provide individual prediction errors at several confidence levels. CCLP computes compound descriptors for a set of input molecules and predicts their activity across the NCI60 panel. The output of running CCLP consists of one barplot per input compound displaying the predicted activities and errors across the NCI60 panel, as well as a text file reporting the predicted activities and errors in prediction\n\nAvailabilityCCLP is freely available on the web at cclp.marseille.inserm.fr

bioinformatics↗

bioRxiv2016

Systematic assessment of multi-gene predictors of pan-cancer cell line sensitivity to drugs exploiting gene expression data

Selected gene mutations are routinely used to guide the selection of cancer drugs for a given patient tumour. Large pharmacogenomic data sets were introduced to discover more of these single-gene markers of drug sensitivity. Very recently, machine learning regression has been used to investigate how well cancer cell line sensitivity to drugs is predicted depending on the type of molecular profile. The latter has revealed that gene expression data is the most predictive profile in the pan-cancer setting. However, no study to date has exploited GDSC data to systematically compare the performance of machine learning models based on multi-gene expression data against that of widely-used single-gene markers based on genomics data.\n\nHere we present this systematic comparison using Random Forest (RF) classifiers exploiting the expression levels of 13,321 genes and an average of 501 tested cell lines per drug. To account for time-dependent batch effects in IC50 measurements, we employ independent test sets generated with more recent GDSC data than that used to train the predictors and show that this is a more realistic validation than K-fold cross-validation. Across 127 GDSC drugs, our results show that the single-gene markers unveiled by the MANOVA analysis tend to achieve higher precision than these RF-based multi-gene models, at the cost of generally having a poor recall (i.e. correctly detecting only a small part of the cell lines sensitive to the drug). Regarding overall classification performance, about two thirds of the drugs are better predicted by multi-gene RF classifiers. Among the drugs with the most predictive of these models, we found pyrimethamine, sunitinib and 17-AAG.

bioinformatics↗