--- title: 'Software Daily Summary: What Happened on October 21, 2026' headline: EMBL-EBI rebuilds Ensembl to handle growing genomic data description: A medicinal-chemistry preprint tests whether machine learning can identify chemists from molecular structures. A clinical development executive argues that AI oversight must cover workflows and accountability. url: https://thebench.bio/news/software/2026-10-21 topic: software date: '2026-10-21' covers: - '2026-10-21' date_published: '2026-10-21T12:05:24Z' date_modified: '2026-10-21T12:05:24Z' tags: - Software - Industry sources: 4 publisher: TheBench.Bio --- # Software Daily Summary: What Happened on October 21, 2026 **EMBL-EBI rebuilds Ensembl to handle growing genomic data** A medicinal-chemistry preprint tests whether machine learning can identify chemists from molecular structures. A clinical development executive argues that AI oversight must cover workflows and accountability. ## Today in brief EMBL-EBI rebuilt Ensembl’s website and technical platform as the resource takes in tens of terabytes of data each month. [1] A preprint from Leash Bio co-founders reports that a model identified chemists from molecular structures, though uneven numbers of molecules per chemist could complicate interpretation. [2] Rajneesh Patil of Syneos Health says clinical AI governance should track how tools affect decisions and who reviews their outputs. [3] [4] ## EMBL-EBI redesigns Ensembl for growing genomic data EMBL-EBI released a redesigned Ensembl website, a free resource for exploring, analyzing, and comparing genomes across thousands of species. [1] Alexey Sokolov, who leads the Genomics Technology Infrastructure Team, said Ensembl receives tens of terabytes of data monthly and the new platform should make scaling easier. [1] The site’s core programming language has changed from Perl to Python, which Sokolov said should simplify future feature development. [1] Ensembl uses genome and transcriptome resources, cross-species information, and AI-based methods to identify genes and interpret genomes. [1] ## Preprint tests whether molecular structures reveal their medicinal chemists Leash Bio co-founders Andrew D. Blevins and Ian K. Quigley published a preprint testing whether machine learning can identify a molecule’s chemist and whether that signal affects activity prediction. [2] The authors reported 27% top-one accuracy and 60% top-five accuracy when identifying a molecule’s chemist from its structure, using data on 1,815 medicinal chemists. [2] In a second experiment, the authors reported that combining chemist-probability values with a protein identifier matched an ECFP4 baseline for predicting binary activity labels. [2] The accompanying discussion cautions that the identification result may be difficult to interpret without knowing how molecules were distributed across chemists. [2] ## Clinical AI governance needs to cover workflows and accountability Rajneesh Patil of Syneos Health argued that connecting AI tools across clinical operations could reduce information handoffs and give teams more time for human judgment. [3] [4] He said faster movement from trial signals to action could help clinical, medical, data, and quality teams address risks earlier. [3] [4] Patil identified three governance weaknesses: treating oversight as a one-time approval, focusing on models instead of workflows and human decisions, and leaving accountability unclear across vendors and systems. [3] [4] He said governance should clarify what AI influenced, who reviewed its output, and how concerns were handled. [3] [4] Patil argued that contract research organizations should coordinate AI systems rather than shift integration work back to sponsors. [3] [4] ## The day in numbers - **Tens of terabytes per month**: Data volume Ensembl receives, according to team lead Alexey Sokolov. [1] - **1,815**: Medicinal chemists included in the preprint’s molecule-author identification task. [2] - **27% and 60%**: Reported top-one and top-five accuracy, respectively, for identifying a molecule’s chemist. [2] ## Also published - [The pre-commercial case for connecting data before acquiring more](), pharmaphorum ## Sources 1. [Welcome: Alexey Sokolov](), EMBL, 2026-10-21 2. [Lessons in Bemistry: identifying the signature of a medicinal chemist](), BLOPIG, 2026-10-21 3. [What Expanding AI Ecosystems Mean for Sponsor Complexity and Oversight](), Applied Clinical Trials, 2026-10-21 4. [Where AI Governance in Clinical Development Falls Short](), Applied Clinical Trials, 2026-10-20 --- TheBench.Bio wrote this from the reporting cited above, with a language model drafting and editing. [1]: [2]: [3]: [4]: