Enhanced NMR-AI Platform able to leverage photo-CIDNP data for Improved Binding Affinity Prediction, Quicker and Finer NMR Structure Determination

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F11992719%3A_____%2F26%3AN0000001" target="_blank" >RIV/11992719:_____/26:N0000001 - isvavai.cz</a>

Výsledek na webu

<a href="https://aiffinity.com/platform/hit-identification" target="_blank" >https://aiffinity.com/platform/hit-identification</a>

DOI - Digital Object Identifier

—

Jazyk výsledku

angličtina

Název v původním jazyce

Enhanced NMR-AI Platform able to leverage photo-CIDNP data for Improved Binding Affinity Prediction, Quicker and Finer NMR Structure Determination

Popis výsledku v původním jazyce

deepHitExplorer is a software result developed within project TM05000031 as an enhanced NMR-AI platform for early-stage fragment-based hit discovery. The software integrates ultrafast 1D photo-CIDNP NMR screening data with artificial intelligence and cheminformatics to improve the identification and prioritisation of small-molecule binders, especially for difficult targets such as intrinsically disordered or highly flexible proteins. The software accepts molecular structures and NMR-derived ligand-epitope information as input and uses a molecular graph attention neural network to predict the probability that new, unseen molecules from large virtual libraries will bind to a target protein. In addition to predicting binding probability, deepHitExplorer estimates which ligand protons are most likely to participate in the binding epitope, thereby providing interpretable atom-level information that can support rational medicinal chemistry decisions, fragment growing and hit optimisation. The software also contains a multi-objective compound-selection module based on a genetic algorithm, which prioritises compounds by predicted binding probability, predicted solubility and structural diversity. A graphical user interface enables users to visualise screened molecules in a 3D UMAP chemical-space plot, inspect predicted ligand epitopes, select candidate compounds and export annotated compound sets for experimental validation. The software was validated in a real drug-discovery case study targeting the androgen receptor splice variant AR-V7. Starting from approximately 480 fragments with high-millimolar affinities, iterative cycles of photo-CIDNP screening, AI model training, virtual screening, compound selection and experimental validation led to the identification of multiple micromolar binders, including compounds in the single-digit micromolar KD range. Cross-validation and enrichment analyses showed that inclusion of NMR-derived epitope vectors improves predictive performance compared with molecule-only models. The result has reached approximately TRL 8, with core functionality implemented, a stable GUI and a deployment-ready version prepared for internal use, partner demonstrations and customer pilots. The software is owned by AI|ffinity s.r.o. and is intended for commercial exploitation through on-premise licensing and through consortium-based research services combining photo-CIDNP NMR, AI-enhanced virtual screening, hit prioritisation and downstream experimental validation. The software and associated know-how are protected as trade secrets and by confidentiality measures while a patent application covering the core concepts is being prepared for filing under the PCT route.

Název v anglickém jazyce

Enhanced NMR-AI Platform able to leverage photo-CIDNP data for Improved Binding Affinity Prediction, Quicker and Finer NMR Structure Determination

Popis výsledku anglicky

deepHitExplorer is a software result developed within project TM05000031 as an enhanced NMR-AI platform for early-stage fragment-based hit discovery. The software integrates ultrafast 1D photo-CIDNP NMR screening data with artificial intelligence and cheminformatics to improve the identification and prioritisation of small-molecule binders, especially for difficult targets such as intrinsically disordered or highly flexible proteins. The software accepts molecular structures and NMR-derived ligand-epitope information as input and uses a molecular graph attention neural network to predict the probability that new, unseen molecules from large virtual libraries will bind to a target protein. In addition to predicting binding probability, deepHitExplorer estimates which ligand protons are most likely to participate in the binding epitope, thereby providing interpretable atom-level information that can support rational medicinal chemistry decisions, fragment growing and hit optimisation. The software also contains a multi-objective compound-selection module based on a genetic algorithm, which prioritises compounds by predicted binding probability, predicted solubility and structural diversity. A graphical user interface enables users to visualise screened molecules in a 3D UMAP chemical-space plot, inspect predicted ligand epitopes, select candidate compounds and export annotated compound sets for experimental validation. The software was validated in a real drug-discovery case study targeting the androgen receptor splice variant AR-V7. Starting from approximately 480 fragments with high-millimolar affinities, iterative cycles of photo-CIDNP screening, AI model training, virtual screening, compound selection and experimental validation led to the identification of multiple micromolar binders, including compounds in the single-digit micromolar KD range. Cross-validation and enrichment analyses showed that inclusion of NMR-derived epitope vectors improves predictive performance compared with molecule-only models. The result has reached approximately TRL 8, with core functionality implemented, a stable GUI and a deployment-ready version prepared for internal use, partner demonstrations and customer pilots. The software is owned by AI|ffinity s.r.o. and is intended for commercial exploitation through on-premise licensing and through consortium-based research services combining photo-CIDNP NMR, AI-enhanced virtual screening, hit prioritisation and downstream experimental validation. The software and associated know-how are protected as trade secrets and by confidentiality measures while a patent application covering the core concepts is being prepared for filing under the PCT route.

Druh

R - Software

CEP obor

—

OECD FORD obor

30401 - Health-related biotechnology

Projekt

<a href="/cs/project/TM05000031" target="_blank" >TM05000031: Zaměření na Androgenní Receptor: Vývoj Vazebných Molekul pro Léčbu Rakoviny Prostaty a Vylepšení NMR-AI Platformy pro Účinnější Návrh Léčiv.</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Rok uplatnění

2026

Kód důvěrnosti údajů

C - Předmět řešení projektu podléhá obchodnímu tajemství (§ 504 Občanského zákoníku), ale název projektu, cíle projektu a u ukončeného nebo zastaveného projektu zhodnocení výsledku řešení projektu (údaje P03, P04, P15, P19, P29, PN8) dodané do CEP, jsou upraveny tak, aby byly zveřejnitelné.

Interní identifikační kód produktu

TM05000031-V2

Technické parametry

deepHitExplorer is a software-only result running on standard computing infrastructure; no dedicated hardware or physical prototype is required. The software uses a molecular graph attention neural network that represents molecules as graphs and incorporates ligand-epitope information derived from photo-CIDNP 1D NMR experiments. It predicts both molecule-level binding probability and atom/proton-level ligand-epitope probabilities. A multi-objective genetic algorithm selects promising compounds from virtual libraries by optimising predicted binding probability, predicted solubility and structural diversity. The graphical interface provides 3D UMAP visualisation of chemical space, inspection of 2D structures and predicted ligand epitopes, and export of selected compounds with annotations to CSV for medicinal chemistry and experimental planning. The software is integrated into the broader NMR-AI platform together with deepScaffOpt and 4D-GRAPHS and is optimised for workflows using photo-CIDNP NMR data and NexMR’s C-bind/Q-bind scoring for flexible and intrinsically disordered protein targets. Verification was performed in the AR-V7 case study through iterative cycles of photo-CIDNP screening, model training, virtual screening, compound selection and experimental validation. Robustness was assessed by repeated 5-fold cross-validation and AUC-ROC, and screening performance by early-recognition metrics such as BEDROC and enrichment measures. The software enriched true binders and enabled progression from high-millimolar fragments to multiple micromolar hits, including more than four compounds in the single-digit micromolar KD range. The result reached approximately TRL 8, with a stable architecture, GUI and deployment-ready version for internal use, partner demonstrations and pilot installations. AI|ffinity s.r.o. owns 100% of the software IP for deepHitExplorer and the broader NMR-AI software modules, while NexMR retains IP related to adapted photo-CIDNP methodology. No concluded third-party licence agreement is reported. Commercial use is planned through AI|ffinity-led licensing, secure on-premise Docker deployment and consortium-based contract research services with AI|ffinity, NexMR and CEITEC MU. A patent filing is being prepared for “Use of NMR Spectroscopy Data for Simultaneous Binding Prediction and Molecular Interaction Localization,” with intended PCT filing. Until filing, the software and know-how are protected by confidentiality regimes, contractual safeguards and trade-secret treatment.

Ekonomické parametry

deepHitExplorer is expected to reduce time and cost in early-stage drug discovery by using ultrafast NMR-derived evidence and AI-based prioritisation to enrich true binders before experimental testing. This reduces the number of compounds that must be purchased, synthesised or tested, improves hit rates, and supports rational medicinal chemistry through interpretable ligand-epitope prediction. The result is a commercializable software asset intended to generate revenue through annual on-premise licences, customer pilot installations and recurring service engagements. It will also serve as the computational backbone of contract research services combining photo-CIDNP screening, AI-enhanced virtual screening, hit prioritisation and NMR-based validation. Implementation is planned for 2026–2035, targeting pharma, biotech, CROs, NMR centres and academic groups working on fragment-based hit discovery, especially for flexible or intrinsically disordered targets.

IČO vlastníka výsledku

11992719

Název vlastníka

AI|ffinity s.r.o.