Brno, Czech Republic and Schlieren, Switzerland—01.05.2025 — AI|ffinity, in collaboration with NexMR, today announced the successful discovery of novel molecular hits against an intrinsically disordered target, the Androgen Receptor isoform 3 (AR-V7), with significantly enhanced binding affinity. This was achieved through the integration of proprietary ultrafast NMR screening technology and advanced AI-driven hit generation.

In this first round of hit discovery, the collaboration yielded molecules exhibiting a 2-log (100-fold) increase in affinity, marking a significant step forward in the partners' joint drug discovery efforts. This achievement marks the first documented integration of photo-CIDNP NMR screening with AI-driven chemical design, showcasing the powerful synergy between the two technologies.

AI and NMR implementation for accelerated drug discovery

“This early result underscores the power of combining cutting-edge machine learning with NMR-based experimental screening,” said Thomas Evangelidis, CEO and CTO of AI|ffinity. “We are excited by the speed and scale at which we’re identifying high-potential hits and believe this collaboration sets a new standard for accelerated therapeutic discovery.”

NexMR AG contributed its proprietary ultrafast NMR screening platform to rapidly evaluate its photohyperpolarizable fragment library against androgen receptor isoform 3 (AR-V7), a protein with a significant intrinsically disordered region, while AI|ffinity applied its advanced AI algorithms to design and optimize candidate molecules and performed the NMR and MST experiments at CEITEC Core Facilities. The project is funded by TAČR programme DELTA2 (project ID: TM05000031).

“We're proud to see our NMR screening platform driving real-world outcomes,” said Matthias Bütikofer, CSO of NexMR AG. “Paired with AI, our technologies are helping push the boundaries of what’s possible in early discovery; no target is out of reach now.”

Targeting Intrinsically Disordered Proteins (IDP)

“Photo-CIDNP not only identifies whether a molecule binds to a protein, but also pinpoints the specific atoms involved in the interaction—without requiring knowledge of the protein’s 3D structure. This makes it particularly well-suited for targeting intrinsically disordered proteins,” said Evangelidis. “We use this interaction data to train our deepHitMiner software, enabling the screen of large chemical libraries and accurately identify compounds likely to bind in a similar fashion by predicting their interacting atoms (ligand epitopes).”

Building on this, he added: “This result - 2 logs of affinity gain in the first round - is exactly the kind of rapid advancement we aimed for. In the next iterations, we will retrain deepHitMiner with the newly acquired photo-CIDNP data, applying a feedback loop optimization strategy to iteratively refine and identify stronger binders in each cycle.”

The partners plan to advance the identified hits into lead compounds, with additional optimization and preclinical studies to follow.