Biology subjects

Elmarakeby, H.

Publications and source records attributed to Elmarakeby, H..

5 recordsLinked to original sources

bioRxiv2026

Expanding P-NET, a multi-purpose biologically informed deep learning framework

We present expanded P-NET, a versatile frame-work for deep learning in computational biology based on P-NET, leveraging biological pathways for interpretable predictions. Our framework achieves competitive performance in genomic & transcriptomic prediction tasks. We demonstrate its stability and interpretability compared to traditional machine learning models. P-NET 2.0 incorporates gene and pathway information, providing valuable insights into complex biological processes. The framework is publicly available, enabling its application to various computational biology tasks.

bioinformatics↗

bioRxiv2025

Empirical Evaluation of Single-Cell Foundation Models for Predicting Cancer Outcomes

Foundation models pretrained on large-scale single-cell RNA sequencing data present a promising opportunity to advance translational cancer research. However, their utility in clinically relevant, patient-level single-cell applications remains underexplored. Here, we developed an agentic strategy to systematically evaluate twelve emerging single-cell foundation models (scFMs) and three alternative baseline approaches across seven cancer-specific tasks, including cell-type annotation, cancer subtype classification, and treatment response prediction. We assessed model performance under zero-shot, continual training, and fine-tuning conditions, conducting 1,530 supervised model-fitting runs and 200 unsupervised subsample evaluations. We found that while current scFMs excelled at certain analysis tasks, such as tumor microenvironment cell annotation, they offered limited advantages in predicting clinical and biological outcomes of cancer patients compared to simpler baseline models. These insights highlight the critical role of scFM evaluation on biologically and clinically relevant tasks for precision oncology. Beyond identifying current limitations, this assessment reveals principles that can guide future methodological innovation and the use of expanded cancer single-cell cohorts to build more biologically informed and translationally effective scFMs. The resulting agentic framework supports the autonomous discovery of emerging scFMs and facilitates their standardized integration and evaluation across cancer-related tasks.

cancer biology↗

bioRxiv2020

Biologically informed deep neural network for prostate cancer classification and discovery

Determination of molecular features that mediate clinically aggressive phenotypes in prostate cancer (PrCa) remains a major biological and clinical challenge. Here, we developed a biologically informed deep learning model (P-NET) to stratify PrCa patients by treatment resistance state and evaluate molecular drivers of treatment resistance for therapeutic targeting through complete model interpretability. Using a molecular cohort of 1,238 prostate cancers, we demonstrated that P-NET can predict cancer state using molecular data that is superior to other modeling approaches. Moreover, the biological interpretability within P-NET revealed established and novel molecularly altered candidates, such as MDM4 and FGFR1, that were implicated in predicting advanced disease and validated in vitro. Broadly, biologically informed fully interpretable neural networks enable preclinical discovery and clinical prediction in prostate cancer and may have general applicability across cancer types.

bioinformatics↗

bioRxiv2020

Clinical interpretation of integrative molecular profiles to guide precision cancer medicine

Individual tumor molecular profiling is routinely used to detect single gene-variant ("first-order") genomic alterations that may inform therapeutic actions -- for instance, a tumor with a BRAF p.V600E variant might be considered for RAF/MEK inhibitor therapy. Interactions between such first-order events (e.g., somatic-germline) and global molecular features (e.g. mutational signatures) are increasingly associated with clinical outcomes, but these "second order" alterations are not yet generally accounted for in clinical interpretation algorithms and knowledge bases. Here, we introduce the Molecular Oncology Almanac (MOAlmanac), a clinical interpretation algorithm paired with a novel underlying knowledge base to enable integrative interpretation of genomic and transcriptional cancer data for point-of-care treatment decision-making and translational hypothesis generation. We compared MOAlmanac to first-order interpretation methodology in multiple retrospective patient cohorts and observed that the inclusion of preclinical and inferential evidence as well as second-order molecular features increased the number of nominated clinical hypotheses. MOAlmanac also performed matchmaking between patient molecular profiles and cancer cell lines to further expand individualized clinical actionability. When applied to a prospective precision oncology trial cohort, MOAlmanac nominated a median of two therapies per patient and identified therapeutic strategies administered in 46% of patient profiles. Overall, we present a novel computational method to perform integrative clinical interpretation of individualized molecular profiles. MOAlmanc increases clinical actionability over conventional approaches by considering second-order molecular features and additional evidence sources, and is available as an open-source framework.

cancer biology↗

bioRxiv2020

Systematic auditing is essential to debiasing machine learning in biology

Representational biases that are common in biological data can inflate prediction performance and confound our understanding of how and what machine learning (ML) models learn from large complicated datasets. However, auditing for these biases is not a common practice in ML in the life sciences. Here, we devise a systematic auditing framework and harness it to audit three different ML applications of significant therapeutic interest: prediction frameworks of protein-protein interactions, drug-target bioactivity, and MHC-peptide binding. Through this, we identify unrecognized biases that hinder the ML process and result in low model generalizability. Ultimately, we show that, when there is insufficient signal in the training data, ML models are likely to learn primarily from representational biases.

systems biology↗