Quantum-Assisted Algorithm QFASG Revolutionises Drug Design, Produces Effective Small Molecule Inhibitors

Researchers from Insilico Medicine Hong Kong Ltd. and Insilico Medicine AI Limited have developed a new algorithm, Quantum-assisted Fragment-based Automated Structure Generator (QFASG), for automated drug design. The algorithm constructs ligands for a target protein using a library of molecular fragments. It was applied to generate new CAMKK2 and ATM inhibitors structures, which are potential targets for cancer treatment. The algorithm successfully designed new low-micromolar inhibitors of CAMKK2 and ATM, highlighting its potential in designing primary hits for further optimization.

Introduction to Quantum-assisted Fragment-based Automated Structure Generator (QFASG)

Automated drug design has seen significant advancements in recent years, with the development of virtual generative models. These models, often powered by deep learning algorithms, have shown promise in producing valuable structures for drug design. However, they often lack interpretability, making understanding the rationale behind their decisions challenging. Furthermore, these algorithms often lack medicinal chemistry expertise, generating chemically insufficient, unstable, toxic, inactive, or previously published structures.

A team of researchers has developed the Quantum-assisted Fragment-based Automated Structure Generator (QFASG) in response to these challenges. This fully automated algorithm is designed to construct ligands for a target protein using a library of molecular fragments. The QFASG was applied to generate new structures of CAMKK2 and ATM inhibitors, resulting in the design of new low-micromolar inhibitors.

The QFASG Pipeline

The QFASG pipeline comprises several interconnected modules. The algorithm can be divided into two parts: the non-iterative section focuses on placing probes (initial fragments) within the target binding site, while the second (iterative) part facilitates fragment-wise structural growth.

Probe placement is conducted through four steps: binding site selection, positioning of probes based on pharmacophore analysis, fine-tuning of probe positions using semiempirical extended tight-binding GFN2-xTB, and final probe selection. The iterative consists of several modules: linking, conformer generation, alignment, rigid docking, and selection.

The algorithm allows several types of fragment linking, such as direct linking, linker-mediated attachment or ring fusing. Once an iteration is completed, the selected structures can be used as starting points for the next iteration until generation stops due to limitations such as molecular weight or exceeding the maximum number of iterations.

Targeting CAMKK2 and ATM for Cancer Treatment

ATM (Ataxia Telangiectasia Mutated) and CAMKK2 (Calcium/Calmodulin-Dependent Protein Kinase Kinase 2) kinases play crucial roles in the pathophysiology of cancer and have emerged as promising targets for cancer treatment. ATM kinase is involved in the DNA damage response and repair, making ATM inhibition a promising approach to sensitize cancer cells. CAMKK2 regulates the activity of CAMK1, CAMK4, and AMP-activated protein kinase (AMPK) and plays a significant role in various physiological and pathological processes, including insulin signalling, metabolic homeostasis, inflammation, and cancer cell growth.

Results and Discussion

To evaluate the capability of QFASG in reproducing known protein ligand structures, a dataset of 17 ligand‒protein complexes was curated. The results showed that QFASG successfully reproduced crystal poses for 7 out of 17 complexes, almost reproduced poses for 8 complexes, and failed to reproduce poses for 2 complexes.

The algorithm was then used to generate structures for new CAMKK2 and ATM inhibitors. Two out of the three compounds synthesized for CAMKK2 exhibited activity in the low micromolar range, while the IC50 value for the third compound exceeded 10 μM. For ATM, one of the three compounds tested exhibited an IC50 value of 4 μM.

Conclusion and Future Directions

The QFASG algorithm has demonstrated its potential in the field of rational drug design, particularly in generating primary hit compound structures for target proteins. The platform’s ability to reproduce the reported crystal structures and to provide novel inhibitors with low-micromolar activity against pharmacologically relevant targets is promising. However, further improvements and refinements are needed to enhance the performance of QFASG, particularly in terms of its accuracy and the diversity of the structures it generates.