Biology subjects

Ester, M.

Publications and source records attributed to Ester, M..

13 recordsLinked to original sources

bioRxiv2024

scMUSCL: Multi-Source Transfer Learning for Clustering scRNA-seq Data

MotivationscRNA-seq analysis relies heavily on single-cell clustering to perform many downstream functions. Several machine learning methods have been proposed to improve the clustering of single cells, yet most of these methods are fully unsupervised and ignore the wealth of publicly available annotated datasets from single-cell experiments. Cells are high-dimensional entities, and unsupervised clustering might find clusters without biological meaning. Exploiting relevant annotated scRNA-seq dataset as the learning reference can provide an algorithm with the knowledge that guides it to better estimate the number of clusters and find meaningful clusters in the target dataset. ResultsIn this paper, we propose Single Cell MUlti-Source CLustering, scMUSCL, a novel transfer learning method for finding clusters of cells in a target dataset by transferring knowledge from multiple annotated source (reference) datasets. scMUSCL relies on a deep neural network to extract domain and batch invariant cell representations, and it effectively addresses discrepancies across multiple source datasets and between source and target datasets in the new representation space. Unlike existing methods, scMUSCL does not need to know the number of clusters in the target dataset in advance and it does not require batch correction between source and target datasets. We conduct extensive experiments using 20 real-life datasets and show that scMUSCL outperforms the existing unsupervised and transfer-learning-based methods in almost all experiments. In particular, we show that scMUSCL outperforms the state-of-the-art transfer-learning-based scRNA-seq clustering method, MARS, by a large margin. AvailabilityThe Python implementation of scMUSCL is available at https://github.com/arashkhoeini/scMUSCL

cell biology↗

bioRxiv2023

Clinical Phenotype Prediction From Single-cell RNA-seq Data using Attention-Based Neural Networks

MotivationA patients disease phenotype can be driven and determined by specific groups of cells whose marker genes are either unknown, or can only be detected at late-stage using conventional bulk assays such as RNA-Seq technology. Recent advances in single-cell RNA sequencing (scRNA-seq) enable gene expression profiling in cell-level resolution, and therefore have the potential to identify those cells driving the disease phenotype even while the number of these cells is small. However, most existing methods rely heavily on accurate cell type detection, and the number of available annotated samples is usually too small for training deep learning predictive models. ResultsHere we propose the method ScRAT for clinical phenotype prediction using scRNA-seq data. To train ScRAT with a limited number of samples of different phenotypes, such as COVID and non-COVID, ScRAT first applies a mixup module to increase the number of training samples. A multi-head attention mechanism is employed to learn the most informative cells for each phenotype without relying on a given cell type annotation. Using three public COVID datasets, we show that ScRAT outperforms other phenotype prediction methods. The performance edge of ScRAT over its competitors increases as the number of training samples decreases, indicating the efficacy of our sample mixup. Critical cell types detected based on high-attention cells also support novel findings in the original papers and the recent literature. This suggests that ScRAT overcomes the challenge of missing marker genes and limited sample number with great potential revealing novel molecular mechanisms and/or therapies.

bioinformatics↗

bioRxiv2022

Ligand Binding Prediction using Protein Structure Graphs and Residual Graph Attention Networks

MotivationComputational prediction of ligand-target interactions is a crucial part of modern drug discovery as it helps to bypass high costs and labor demands of in vitro and in vivo screening. As the wealth of bioactivity data accumulates, it provides opportunities for the development of deep learning (DL) models with increasing predictive powers. Conventionally, such models were either limited to the use of very simplified representations of proteins or ineffective voxelization of their 3D structures. Herein, we present the development of the PSG-BAR (Protein Structure Graph -Binding Affinity Regression) approach that utilizes 3D structural information of the proteins along with 2D graph representations of ligands. The method also introduces attention scores to selectively weight protein regions that are most important for ligand binding. ResultsThe developed approach demonstrates the state-of-the-art performance on several binding affinity benchmarking datasets. The attention-based pooling of protein graphs enables identification of surface residues as critical residues for protein-ligand binding. Finally, we validate our model predictions against an experimental assay on a viral main protease (Mpro)- the hallmark target of SARS-CoV-2 coronavirus. AvailabilityThe code for PSG-BAR is made available at https://github.com/diamondspark/PSG-BAR Contactacherkasov@prostatecentre.com

bioinformatics↗

bioRxiv2021

Velodrome: Out-of-Distribution Generalization from Labeled and Unlabeled Gene Expression Data for Drug Response Prediction

Data discrepancy between preclinical and clinical datasets poses a major challenge for accurate drug response prediction based on gene expression data. Different methods of transfer learning have been proposed to address this data discrepancy. These methods generally use cell lines as source domains and patients, patient-derived xenografts, or other cell lines as target domains. However, they assume that they have access to the target domain during training or fine-tuning and they can only take labeled source domains as input. The former is a strong assumption that is not satisfied during deployment of these models in the clinic. The latter means these methods rely on labeled source domains which are of limited size. To avoid these assumptions, we formulate drug response prediction as an out-of-distribution generalization problem which does not assume that the target domain is accessible during training. Moreover, to exploit unlabeled source domain data, which tends to be much more plentiful than labeled data, we adopt a semi-supervised approach. We propose Velodrome, a semi-supervised method of out-of-distribution generalization that takes labeled and unlabeled data from different resources as input and makes generalizable predictions. Velodrome achieves this goal by introducing an objective function that combines a supervised loss for accurate prediction, an alignment loss for generalization, and a consistency loss to incorporate unlabeled samples. Our experimental results demonstrate that Velodrome outperforms state-of-the-art pharmacogenomics and transfer learning baselines on cell lines, patient-derived xenografts, and patients. Finally, we showed that Velodrome models generalize to different tissue types that were well-represented, under-represented, or completely absent in the training data. Overall, our results suggest that Velodrome may guide precision oncology more accurately.

bioinformatics↗

bioRxiv2021

Drug Sensitivity Prediction From Cell Line-Based Pharmacogenomics Data: Guidelines for Developing Machine Learning Models

The goal of precision oncology is to tailor treatment for patients individually using the genomic profile of their tumors. Pharmacogenomics datasets such as cancer cell lines are among the most valuable resources for drug sensitivity prediction, a crucial task of precision oncology. Machine learning methods have been employed to predict drug sensitivity based on the multiple omics data available for large panels of cancer cell lines. However, there are no comprehensive guidelines on how to properly train and validate such machine learning models for drug sensitivity prediction. In this paper, we introduce a set of guidelines for different aspects of training gene expression-based predictors using cell line datasets. These guidelines provide extensive analysis of the generalization of drug sensitivity predictors, and challenge many current practices in the community including the choice of training dataset and measure of drug sensitivity. Application of these guidelines in future studies will enable the development of more robust preclinical biomarkers.

bioinformatics↗

bioRxiv2020

Identification of Differentially Expressed Gene Modules in Heterogeneous Diseases

MotivationIdentification of differentially expressed genes is necessary for unraveling disease pathogenesis. This task is complicated by the fact that many diseases are heterogeneous at the molecular level and samples representing distinct disease subtypes may demonstrate different patterns of dysregulation. Biclustering methods are capable of identifying genes that follow a similar expression pattern only in a subset of samples and hence can consider disease heterogeneity. However, identifying biologically significant and reproducible sets of genes and samples remains challenging for the existing tools. Many recent studies have shown that the integration of gene expression and protein interaction data improves the robustness of prediction and classification and advances biomarker discovery. ResultsHere we present DESMOND, a new method for identification of Differentially ExpreSsed gene MOdules iN Diseases. DESMOND performs network-constrained biclustering on gene expression data and identifies gene modules -- connected sets of genes up- or down-regulated in subsets of samples. We applied DESMOND on expression profiles of samples from two large breast cancer cohorts and have shown that the capability of DESMOND to incorporate protein interactions allows identifying the biologically meaningful gene and sample subsets and improves the reproducibility of the results. Availabilityhttps://github.com/ozolotareva/DESMOND Contactozolotareva@techfak.uni-bielefeld.de Supplementary informationSupplementary data are available at Bioinformatics online.

bioinformatics↗

bioRxiv2020

AITL: Adversarial Inductive Transfer Learning with input and output space adaptation for pharmacogenomics

MotivationThe goal of pharmacogenomics is to predict drug response in patients using their single- or multi-omics data. A major challenge is that clinical data (i.e. patients) with drug response outcome is very limited, creating a need for transfer learning to bridge the gap between large pre-clinical pharmacogenomics datasets (e.g. cancer cell lines), as a source domain, and clinical datasets as a target domain. Two major discrepancies exist between pre-clinical and clinical datasets: 1) in the input space, the gene expression data due to difference in the basic biology, and 2) in the output space, the different measures of the drug response. Therefore, training a computational model on cell lines and testing it on patients violates the i.i.d assumption that train and test data are from the same distribution. ResultsWe propose Adversarial Inductive Transfer Learning (AITL), a deep neural network method for addressing discrepancies in input and output space between the pre-clinical and clinical datasets. AITL takes gene expression of patients and cell lines as the input, employs adversarial domain adaptation and multi-task learning to address these discrepancies, and predicts the drug response as the output. To the best of our knowledge, AITL is the first adversarial inductive transfer learning method to address both input and output discrepancies. Experimental results indicate that AITL outperforms state-of-the-art pharmacogenomics and transfer learning baselines and may guide precision oncology more accurately. Availability of codes and supplementary materialhttps://github.com/hosseinshn/AITL Contactccollins@prostatecentre.com and ester@cs.sfu.ca

bioinformatics↗

bioRxiv2020

Deep Learning Modelling of Androgen Receptor Responses to Prostate Cancer Therapies

Gain-of-function mutations in human Androgen Receptor (AR) are amongst major causes of drug resistance in prostate cancer (PCa). Identifying mutations that cause resistant phenotype is of critical importance for guiding treatment protocols as well as for designing drugs that do not elicit adverse responses. However, experimental characterization of these mutations is time consuming and costly, and therefore predictive models are needed to anticipate resistant mutations and to guide drug discovery process. In this work, we leverage experimental data collected on 69 clinically observed and/or literature described AR mutants to train a deep neural network (DNN) to predict their responses to currently used and experimental AR anti-androgens. We demonstrate that the use of DNN provides more accurate prediction of the biological outcome (inhibition, activation, no-response) in AR mutant-drug pairs compared to other machine learning approaches and also allows the use of more general 2D descriptors. Finally, the developed approach was used to predict the effect of the latest AR inhibitor darolutamide on all reported AR mutants.

cancer biology↗

bioRxiv2019

BDKANN - Biological Domain Knowledge-based Artificial Neural Network for drug response prediction

MotivationOne of the main goals of precision oncology is to predict the response of a patient to a given cancer treatment based on their genomic profile. Although current models for drug response prediction are becoming more accurate, they are also black boxes and cannot explain their predictions, which is of particular importance in cancer treatment. Many models also do not leverage prior biological knowledge, such as the hierarchical information on how proteins form complexes and act together in pathways. ResultsIn this work, we use this prior biological knowledge to form the architecture of a deep neural network to predict cancer drug response from cell line gene expression data. We find that our approach not only has a low prediction error compared to baseline models but also allows meaningful interpretation of the network. These interpretations can both explain predictions made and discover novel connections in the biological knowledge that may lead to new hypotheses about mechanisms of drug action. AvailabilityCode at https://github.com/osnow/BDKANN Supplementary informationIncluded with submission

bioinformatics↗

bioRxiv2019

Uncovering the subtype-specific temporal order of cancer pathway dysregulation

Cancer is driven by genetic mutations that dysregulate pathways important for proper cell function. Therefore, discovering these cancer pathways and their dysregulation order is key to understanding and treating cancer. However, the heterogeneity of mutations between different individuals makes this challenging and requires that cancer progression is studied in a subtype-specific way. To address this challenge, we provide a mathematical model, called Subtype-specific Pathway Linear Progression Model (SPM), that simultaneously captures cancer subtypes and pathways and order of dysregulation of the pathways within each subtype. Experiments with synthetic data indicate the robustness of SPM to problem specifics including noise compared to an existing method. Moreover, experimental results on glioblastoma multiforme and colorectal adenocarcinoma show the consistency of SPMs results with the existing knowledge and its superiority to an existing method in certain cases. The implementation of our method is available at https://github.com/Dalton386/SPM.

bioinformatics↗

bioRxiv2019

SUBSTRA: Supervised Bayesian Patient Stratification

Patient stratification methods are key to the vision of precision medicine. Here, we consider transcriptional data to segment the patient population into subsets relevant to a given phenotype. Whereas most existing patient stratification methods focus either on predictive performance or interpretable features, we developed a method striking a balance between these two important goals. We introduce a Bayesian method called SUBSTRA that uses regularized biclustering to identify patient subtypes and interpretable subtype-specific transcript clusters. The method iteratively re-weights feature importance to optimize phenotype prediction performance by producing more phenotype-relevant patient subtypes. We investigate the performance of SUBSTRA in finding relevant features using simulated data and successfully benchmark it against state-of-the-art unsupervised stratification methods and supervised alternatives. Moreover, SUBSTRA achieves predictive performance competitive with supervised benchmark methods and provides interpretable transcriptional features in diverse biological settings, such as drug response prediction, cancer diagnosis, or kidney transplant rejection.

bioinformatics↗

bioRxiv2019

MOLI: Multi-Omics Late Integration with deep neural networks for drug response prediction

MotivationHistorically, gene expression has been shown to be the most informative data for drug response prediction. Recent evidence suggests that integrating additional omics can improve the prediction accuracy which raises the question of how to integrate the additional omics. Regardless of the integration strategy, clinical utility and translatability are crucial. Thus, we reasoned a multi-omics approach combined with clinical datasets would improve drug response prediction and clinical relevance. ResultsWe propose MOLI, a Multi-Omics Late Integration method based on deep neural networks. MOLI takes somatic mutation, copy number aberration, and gene expression data as input, and integrates them for drug response prediction. MOLI uses type-specific encoding subnetworks to learn features for each omics type, concatenates them into one representation and optimizes this representation via a combined cost function consisting of a triplet loss and a binary cross-entropy loss. The former makes the representations of responder samples more similar to each and different from the non-responders, and the latter makes this representation predictive of the response values. We validate MOLI on in vitro and in vivo datasets for five chemotherapy agents and two targeted therapeutics. Compared to state-of-the-art single-omics and early integration multi-omics methods, MOLI achieves higher prediction accuracy in external validations. Moreover, a significant improvement in MOLIs performance is observed for targeted drugs when training on a pan-drug input, i.e. using all the drugs with the same target compared to training only on drug-specific inputs. MOLIs high predictive power suggests it may have utility in precision oncology. Availability of the implemented codeshttps://github.com/hosseinshn/MOLI Contactccollins@prostatecentre.com and ester@cs.sfu.ca

bioinformatics↗

bioRxiv2018

Deep Genomic Signature for early metastasis prediction in prostate cancer

For prostate cancer patients, timing and intensity of therapy are adjusted based on their prognosis. Clinical and pathological factors, and recently, gene expression-based signatures have been shown to predict metastatic prostate cancer. Previous studies used labelled datasets, i.e. those with information on the metastasis outcome, to discover gene signatures to predict metastasis. Due to steady progression of prostate cancer, datasets for this cancer have a limited number of labelled samples but more unlabelled samples. In addition to this issue, the high dimensionality of the gene expression data also poses a significant challenge to train a classifier and predict metastasis accurately. In this study, we aim to boost the prediction accuracy by utilizing both labelled and unlabelled datasets together. We propose Deep Genomic Signature (DGS), a method based on Denoising Auto-Encoders (DAEs) and transfer learning. DGS has the following steps: first, we train a DAE on a large unlabelled gene expression dataset to extract the most salient features of its samples. Then, we train another DAE on a small labelled dataset for a similar purpose. Since the labelled dataset is small, we employ a transfer learning approach and use the parameters learned from the first DAE in the second one. This approach enables us to train a large DAE on a small dataset. After training the second DAE, we obtain the list of genes with high weights by applying a standard deviation filter on the transferred and learned weights. Finally, we train an elastic net logistic regression model on the expression of the selected genes to predict metastasis. Because of the elastic net regularization, some of the selected genes have non-zero coefficients in the classifier which we consider as the DGS gene signature for metastasis. We apply DGS to six labelled and one large unlabelled prostate cancer datasets. Results on five validation datasets indicate that DGS outperforms state-of-the-art gene signatures (obtained from only labelled datasets) in terms of prediction accuracy. Survival analyses demonstrate the potential clinical utility of our gene signature that adds novel prognostic information to the well-established clinical factors and the state-of-the-art gene signatures. Finally, pathway analysis reveals that the DGS gene signature captures the hallmarks of prostate cancer metastasis. These results suggest that our method helps to identify a robust gene signature that may improve patient management.

bioinformatics↗