In-silico Drug Discovery: Companies, Startups & Research

Notable company participants include Astrazeneca, InSilico Medicine, Microsoft, Pfizer and Hoffmann-La Roche. Their work spans AI-assisted molecular design, computational target discovery, binding prediction and cheminformatics. Company activity varies: some organizations develop platforms, while others apply computational methods within broader discovery programs. These examples are drawn from organizations identified in the topic data.

Iktos, MindRank AI Ltd. and Qubit Pharmaceuticals SAS are notable startups in the data for 2026. Iktos combines generative AI with retrosynthesis and laboratory robotics. MindRank AI Ltd. works on molecular-dynamics refinement for affinity prediction, while Qubit Pharmaceuticals SAS explores quantum-computing approaches and small-molecule binding to RNA. The data identifies 12 startups overall, but does not rank them by promise.

Chinese Academy of Sciences CAS (512 window documents), King Saud University (469) and Zhejiang University (320) are among the leading research organizations in the activity data. Their recent work includes structure-based virtual screening, free-energy perturbation combined with machine learning, and deep-learning prediction of protein–ligand binding affinity. The document figures describe activity in the topic window, not a measure of research quality.

Recorded grant funding totals $651,895,741 across 618 grants since 2020. The data also lists 28 funding rounds totaling $114,257,193,831. That round figure should be interpreted carefully: the entries include corporate expansions and infrastructure finance, not only investment in in-silico discovery companies. The grant total is the more direct indicator of public and research funding represented in these figures.

No: patent activity declined 54.1% in the last 12 months versus the previous 12 months, while the 24-month window contains 124 patents. This contrasts with publication activity, which grew 35.4% to 15,504 in the last year. The difference indicates that recent research output has strengthened even as patent activity has fallen; it does not, by itself, establish why.

Recent examples include DrugCLIP (August 2026), New York University’s tautomer-predictor AI (September 2026), and an audit of 51 benchmark datasets (August 2026). DrugCLIP applies contrastive learning to genome-wide virtual screening, while the NYU model predicts hydrogen placement associated with molecular stability. The benchmark audit raises a separate concern: dataset problems can affect how AI methods compare.

The main approaches listed are structure-based design, ligand-based modeling, physics-based simulation, AI and generative chemistry, and computational target discovery. Structure-based methods use target structures; ligand-based methods learn from known compounds. Physics-based simulation models molecular interactions, AI predicts or generates molecules, and computational target discovery prioritizes biological targets. Each approach has distinct data, modeling and validation requirements.

The United States leads the activity data with 51,266 documents (41.5%), followed by China with 26,410 (21.4%). The United Kingdom and India also have substantial activity, with 20,253 and 19,750 documents respectively. These country figures describe document associations in the data; they should not be read as exclusive shares of global activity.

Oncology is the largest listed application area, with 17,019 documents; infectious diseases follows with 5,810. Other listed areas include neurological disorders, with 4,814 documents, and cardiometabolic disease, with 4,501. Computational methods can support target selection, virtual screening and lead optimization across these fields, but the document totals indicate activity rather than clinical effectiveness.