This landscape covers patent records that combine core computational toxicology methods — read-across, QSAR modelling and in silico toxicology — with the supporting technical work needed to make those methods defensible, such as analogue identification, data gap filling, uncertainty documentation, applicability domain definition and structural alert flagging. It excludes records that only mention these terms in passing without claiming the underlying method. The scope runs from 2015 through mid-2026 across 104 published records.

The ranking spans 48 companies, with the leading assignee holding 9 of the 104 records and the fifth-ranked holder at 8. The top 5 assignees together account for 42.3% of all records, and the top 10 account for 65.4%, which means the field has meaningful concentration at the top but still leaves a long tail of smaller filers. No single company holds anything close to a majority of filings.

Filing activity peaked at 9 records in 2021 and returned to 9 records again in 2024, a 0% change across that three-year span — best read as a plateau rather than growth or decline. Figures for 2025 and 2026 are still incomplete because patent publication typically lags filing by around 18 months, so recent-looking dips should not be read as a real slowdown yet. 2024 is the most recent year that can be treated as reasonably complete.

The most-cited records in this landscape concern machine-learning-based cardiotoxicity prediction and computer-aided compound design, with the top-cited record drawing 65 citations. These older, heavily cited filings define much of the prior art that later machine-learning toxicity prediction claims need to design around. Newer filings, such as the 2025 chemical toxicity prediction record from National Taiwan University, build on QSAR modelling in ways that should be checked against this citation base.

Based on the IPC composition, claim density concentrates in digital data processing (G06F, 41.3% of records) and computational chemistry (G16C, 34.6%), while the documentation and validation side of read-across work — uncertainty documentation, applicability domain scoring and structural alert validation — shows up in the search scope but with thinner claim coverage. That gap suggests more room to file defensible claims around the evidentiary and validation workflow than around core QSAR modelling itself, where prior art is denser.