Australian researchers, led by Monash University, have invented a new artificial intelligence (AI) tool poised to reshape virtual screening in early-stage drug discovery and enhance scientists' ability to identify potential new medicines. The Australian invention 'PSICHIC' (PhySIcoCHemICal) brings together expertise at the interface of computing technology and drug discovery to offer an entirely new approach.

According to the researchers, the study demonstrates how PSICHIC uses only sequence data and AI to decode protein-molecule interactions with state-of-the-art accuracy while eliminating the need for costly and less accurate processes such as 3D structures.

According to Dr Lauren May, co-lead author from the Monash Institute of Pharmaceutical Sciences (MIPS), a comparison of experimental and AI predictions of a large compound library against the A1 receptor - a potential therapeutic target for many diseases - demonstrated PSICHIC could effectively screen and identify a novel drug candidate. Moreover, PSICHIC was able to distinguish the functional effects of the compound or, in other words, how the drug might affect our bodies.

Dr May stated that the team has already demonstrated that PSICHIC can effectively screen new drug candidates and perform selectivity profiling. She believes AI has enormous potential to completely change the drug discovery landscape.

PSICHIC and drug discovery

Determining the binding affinity and functional effects of small-molecule ligands on proteins is critical in drug discovery. Current computational methods can predict these protein–ligand interaction properties but often lose accuracy without high-resolution protein structures and falter in predicting functional effects. 

PSICHIC incorporates physicochemical constraints to decode interaction fingerprints directly from sequence data alone. This enables PSICHIC to attain capabilities in decoding mechanisms underlying protein–ligand interactions, achieving state-of-the-art accuracy and interpretability. 

Trained on identical protein–ligand pairs without structural data, PSICHIC matched and even surpassed leading structure-based methods in binding-affinity prediction. In an experimental library screening for adenosine A1 receptor agonists, PSICHIC discerned functional effects effectively, ranking the sole novel agonist within the top three.

PSICHIC's interpretable fingerprints identified protein residues and ligand atoms involved in interactions, and helped in unveiling selectivity determinants of protein–ligand interaction.

Data scientist, AI expert, and lead author Professor Geoff Webb from Monash's Department of Data Science and Artificial Intelligence said that while other methods for predicting protein-molecule interactions already exist, they can be expensive and fail to predict a drug's functional effects.

According to Professor Webb, the application of AI approaches to enhance the affordability and accuracy of drug discovery is a rapidly expanding area. With PSICHIC, their team has eliminated the need for 3D structures to map protein-molecule interactions, which is a costly and often restrictive requirement. 

Core building block

Dr. Anh Nguyen, co-lead author from MIPS with strong expertise in AI approaches to drug-receptor interactions, emphasised the importance of these interactions by stating that interaction between molecules and proteins underpin many biological processes, with drugs exerting their intended effects by selectively interacting with specific proteins. 

Significant global efforts have been to develop new AI-based methods to accurately determine how a molecule might behave when interacting with its protein target. In her opinion, this is the "core building block" to making medicines.

The researchers remarked that AI has the potential to dramatically improve the robustness, efficiency and cost at multiple stages during the drug discovery process, from early stage discoveries right through to predicting clinical responses. 

However, since many AI systems fundamentally rely on pattern matching, these systems can suffer from unrestrained degrees of freedom. This can lead to memorisation of previously known patterns rather than learning the underlying mechanisms of protein-ligand interactions, ultimately hindering the discovery of novel drugs.

Monash University