OpenRounds Editorial
Daily Briefing
Monday, July 20, 2026
What Changed
A Bioresource Technology study surveys AI-driven secretion leader engineering in yeast, Samsung Biologics bids for a peptide manufacturer, and Epic describes ambient scribing reaching roughly 90% note accuracy from a single recording [1][2][3].
Industry & Products
•[AI in Clinical Operations] Epic's ambient scribe captures orders, diagnoses, and documentation from a single visit recording, delivering notes at roughly 90% accuracy so clinicians fine-tune rather than draft from scratch [1]. The vendor-reported figure has no independent validation, but the structured visit capture — pulling diagnoses and orders alongside narrative — marks a shift from passive transcription toward encounter-level data generation.
•[AI in Biopharma] Samsung Biologics has bid to acquire a peptide manufacturer, intending to embed an in-house AI-driven peptide design platform and boost contract manufacturing capacity [3]. CDMO buyers evaluating peptide development partners should watch whether combined computational design and scaled production compress timelines enough to shift vendor selection criteria.
•[AI in Biopharma] A Bioresource Technology study surveys AI-driven secretion leader engineering strategies for yeast recombinant protein production, covering rational modification, hybrid leader design, codon optimization, and fusion approaches [2]. Bioprocessing R&D groups get a methodology landscape for accelerating secretion-efficiency optimization, not a validated commercial tool.
Policy & Ops
•[AI in Clinical Policy] FDA's July 2026 draft guidance requires AI/ML device manufacturers to demonstrate HIPAA-compliant data de-identification in pre-market submissions, forcing developers to reconcile patent disclosure, regulatory approval, and privacy obligations simultaneously [4]. Regulatory leads should expect pre-market data packages to expand beyond model performance into provable de-identification protocols.
•[AI in Clinical Operations] The British Society of Echocardiography issued a consensus position on AI integration in echo services, addressing governance and equitable access alongside clinical validation [5]. Cardiology imaging directors gain a specialty-society template for structured AI adoption criteria — a reference framework usable even outside the UK.
Research
•[AI in Clinical Operations] Researchers in npj Digital Medicine propose machine unlearning as a governance requirement for clinical AI, arguing that removing learned information from deployed models must be auditable and clinically safe before updates can be trusted [6]. Clinical AI governance teams get an early framework for update accountability, though operational methods for verifying what a model has forgotten remain nascent.
•[AI Evidence] Spec-ReX, a causal-responsibility explainability method for vibrational spectroscopy, outperformed SHAP and Grad-CAM on ground-truth localization in oral FTIR and esophageal Raman diagnostics [7]. Current XAI tools often highlight irrelevant spectral regions, and Spec-ReX's IoU of 0.10 — while the best reported — still leaves substantial localization error before clinical deployment.
One to Watch
•[AI in Clinical Practice] A UK-based trial is testing AI-powered atrial fibrillation detection and prediction using consumer wearables, targeting a condition affecting over one million people in the UK with significant morbidity and cost [8]. The trial is no longer recruiting and results are not yet reported, but detection performance could anchor a procurement case for wearable-based AF screening.