6 April 2023 in Expert Systems with Applications
18 January 2025 in Database: The Journal of Biological Databases and Curation
GGGraciela Gonzalez-Hernandez
Graciela Gonzalez-Hernandez
+27
We present here an overview of the BioCreative VIII Task 3 competition, which called for the development of state-of-the-art approaches to automatic normalization of observations noted by physicians in dysmorphology physical examinations to the Human Phenotype Ontology (HPO). We made available for the task 3136 deidentified and manually annotated observations extracted from electronic health records of 1652 paediatric patients at the Children’s Hospital of Philadelphia. This task is challenging due to the discontinuous, overlapping, and descriptive mentions of the observations corresponding to HPO terms, severely limiting the performance of straightforward strict matching approaches. Ultimately, an effective automated solution to the task will facilitate computational analysis that could uncover novel correlations and patterns of observations in patients with rare genetic diseases, enhance our understanding of known genetic conditions, and even identify previously unrecognized conditions. A total of 20 teams registered, and 5 teams submitted their predictions. We summarize the corpus, the competing systems approaches, and their results. The top system used a pre-trained large language model and achieved a 0.82 F1 score, which is close to human performance, confirming the impact that recent advances in natural language processing can have on tasks such as this. The post-evaluation period of the challenge, at https://codalab.lisn.upsaclay.fr/competitions/11351, will be open for submissions for at least 18 months past the end of the competition. Database URL: https://codalab.lisn.upsaclay.fr/competitions/11351
1 July 2026 in Bioinformatics
: Identifying metabolic sites where cytochrome P450 enzymes metabolize small-molecule drugs is essential for drug discovery. Although existing computational approaches have been proposed for site-of-metabolism prediction, they typically ignore cytochrome P450 isoform identity or model isoforms independently, thereby failing to fully capture inherent cross-isoform metabolic patterns. In addition, prior evaluations often rely on top-k metrics, where false positive atoms may be included among the top predictions, underscoring the need for complementary metrics that more directly assess binary atom-level discrimination under severe class imbalance. We propose ATTNSOM, an atom-level site-of-metabolism prediction framework that integrates intrinsic molecular reactivity with cross-isoform relationships. The model combines a shared graph encoder, molecule-conditioned atom representations, and a cross-attention mechanism to capture correlated metabolic patterns across cytochrome P450 isoforms. The model is evaluated on two benchmark datasets annotated with site-of-metabolism labels at atom resolution. Across these benchmarks, the model achieves consistently strong top-k performance across multiple cytochrome P450 isoforms. Relative to ablated variants, the model yields higher Matthews correlation coefficient, indicating improved discrimination of true metabolic sites. These results support the importance of explicitly modeling cross-isoform relationships for site-of-metabolism prediction. : The code and datasets are available at https://github.com/dmis-lab/ATTNSOM.
26 December 2024 in Bioinformatics Advances
Motivation: Understanding which genes are significantly affected by drugs is crucial for drug repurposing, as drugs targeting specific pathways in one disease might be effective in another with similar genetic profiles. By analyzing gene expression changes in cells before and after drug treatment, we can identify the genes most impacted by drugs. Results: The Biologically-Aware Interpretable Differential Gene Expression Ranking (BADGER) model is an interpretable model designed to predict gene expression changes resulting from interactions between cancer cell lines and chemical compounds. The model enhances explainability through integration of prior knowledge about drug targets via pathway information, handles novel cancer cell lines through similarity-based embedding, and employs three attention blocks that mimic the cascading effects of chemical compounds. This model overcomes previous limitations of cell line range and explainability constraints in drug–cell response studies. The model demonstrates superior performance over baselines in both unseen cell and unseen pair split evaluations, showing robust prediction capabilities for untested drug–cell line combinations. Availability and implementation: This makes it particularly valuable for drug repurposing scenarios, especially in developing therapeutic plans for new or resistant diseases by leveraging similarities with other diseases. All code and data used in this study are available at https://github.com/dmis-lab/BADGER.git.