Real deployments and the repeatable workflow patterns behind them — what teams are running in discovery, development, manufacturing, and the clinic, and how each workflow actually hangs together.
Use cases are single workflows, described end to end — either a named deployment at a specific organization, or an anonymized workflow pattern that teams run in practice. Guides, playbooks, and cross-cutting analysis live in Articles. Vendor/service provider-authored case studies about their own product or service are commercial content and live with that listing in Products & Services.
Paige reports a sequential reader study of AI-assisted prostate biopsy review, documenting clinical evaluation associated with FDA De Novo authorization.
Bayer tested explainable ML to support drug-safety signal classification, comparing predictions with safety experts in a three-month prospective evaluation.
Models trained on feedback from 35 Novartis chemists were tested for compound prioritization and molecular design.
Novartis and MIT evaluated ML models predicting drug approvals from clinical trial data. The best model achieved 0.88 AUC versus a 0.78 baseline.
Moderna used a generative neural network and automated protein engineering to identify OTC enzyme variants with improved activity and thermal stability.
Researchers apply automated ML model selection to two antibody manufacturing datasets to predict quality attributes and identify process drivers.
Genentech use case: Lab-in-the-loop combines AI, active learning, and lab testing to optimize antibodies, delivering 3-100× better binding variants for therapeutic targets.
Genentech details DyAb, a pairwise model predicting binding affinity changes in low-data regimes, optimizing EGFR and IL-6 antibodies with up to 10-fold affinity gains.
An operational retrospective examining Pfizer’s multi-stage AI implementation.
How formulation and process teams combine mechanistic freeze-drying and spray-drying models with data-driven methods to map the design space — shelf temperature, chamber pressure, primary-drying endpoint, collapse limits, residual moisture — before committing scarce material to a cycle.
How teams use active learning, Bayesian optimization, and design-of-experiments models to navigate the large combinatorial space of cell-free reaction components — lysate, energy system, cofactors, salts, temperature — and reach higher yields with far fewer benchtop reactions.
How trial teams use models on clinical datasets once enrolment is underway — data-quality monitoring, query reduction, adverse-event coding, and narrative drafting — with humans making every clinical and regulatory decision.
What AI workflow would you like to see broken down? We research and write the most requested patterns.
