Integrating AI and HCI for drug discovery
Posted 1st October 2018 by Kieran Chambers
Bringing new therapeutics to patients is easier said than done due to the enormous intrinsic complexity behind being efficacious and safe.
Drug design: an intrinsic multi-objective optimisation problem
For a compound to be efficacious in the treatment of a specific disease it should have, in primis, a good biological activity towards the target (i.e. generally an enzyme, receptor, ion channel or transporter) triggering a therapeutic effect on the disease. Beside this constitutional activity, to be efficacious the compound must be absorbed from the site of administration, get to the systemic circulation and from there it should be distributed along the body in order to reach the site of action (i.e. the specific organ, tissue, cell and sub-cellular location where the therapeutic target resides). During this trip, the compound has to maintain its active form without being metabolised and/or eliminated ahead of time.
For a compound to be safe, once it has promoted its therapeutic action on the target, it should be metabolised and excreted without it or any of its metabolites causing any toxic effect (e.g. hepatotoxicity, cardiotoxicity, cytotoxicity, etc). Additionally, it should avoid any direct or indirect interaction with other therapeutics that, eventually, may be co-administrated (drug-drug interaction).
Each of the aforementioned necessities is controlled and regulated by distinct physical, chemical and biological processes that, in turn, depend on different molecular properties. These properties can be tuned by modifying the chemical structure of the compound but unfortunately often the optimisation of one property leads to the worsening of another. For this reason drug design is an intrinsic multi-objective optimisation problem where several objectives must be taken into account and optimised at once.
AI: a versatile player in drug design
Artificial Intelligence has been successfully adopted in drug design. It can help design new promising molecules (i.e. de novo drug design), to find the most efficient synthesis routes, to predict the chemical stability, the biological activity, the bioavailability and other essential pharmacokinetics properties. Nowadays it constitutes an additional player with diverse capabilities in the drug design field.
The need for many specialists
The different disciplines involved in drug design are mastered by distinct professionals. For example: synthetic chemists assess the chemical stability and synthetic accessibility of molecules; medicinal chemists evaluate the biological activity and pharmacokinetics; computational chemists use in silico approaches to simulate the molecular interaction of the molecule with the target; toxicologists may give insight on how to minimize possible toxicity problems; etc. Each expert does not only tackle the problem from her/his professional perspective and professional knowledge but also contributes with her/his own expertise, personal experience and individual problem solving skill.
Human Collective Intelligence (HCI) is a shared intelligence, an emergent property arising from the collaboration of several individuals. These individuals establish synergistic interactions so that the final contribution to the solution of a certain problem is larger than the contribution of each. The collective intelligence appears, inter alia, in the context of consensus decision making and crowdsourcing applications. HCI has not yet been thoughtfully explored in drug design, although such an approach has proven powerful in other fields. Individual contributions of specialists in form of molecule evaluations, design ideas and molecule optimisation strategies can provide in a collaborative environment completely new dynamics for the development of drug design.
Unique Therapeutics Through “Consensus Design And Decision Taking”
One problem of current drug design is that it is a fragmented process where each player (i.e. specialists and AI) works individually and somehow in isolation from the others. Each player evaluates and modifies the molecules following her/his/its own criteria, often disregarding the evaluation and proposals done by the other players. In this fragmented process only a limited number of design ideas and evaluations are considered and often suboptimal molecules are chosen. As a result, unique therapeutic opportunities are potentially lost on the way.
At Molomics we believe that the integration of AI with Human Intelligence and its advanced form of HCI in a single, transparent drug design process will allow to find unique therapeutics. Thereby, both AI and HCI perform activities of molecule design & evaluation.
AI can design millions of new molecules and optimise existing ones in a high throughput mode using de novo methods. HCI follows principally an atom-based molecule design strategy led by the experience and intuition of the team. AI can evaluate molecules with a multi-objective strategy using many predictive methods for different endpoints (e.g. pharmacological properties). HCI evaluates molecules using the team’s knowledge and expertise to assess and adjust AI evaluations and, additionally, it can score other features of the designed molecules, which cannot be determined by AI (e.g. synthetic accessibility). Eventually, these human originated scorings can also be reused by AI.
In order to get the most from these two complementary worlds it is important that the results of AI (i.e. molecules and evaluations) are directly made available to HCI and those of HCI to AI. In this iterative procedure the “final” results of a player can become the starting points of the other. Thereby each intelligence learns from and takes advantage of the other. From a theoretical perspective, this process can be pictured as a genetic algorithm where the genes coming from AI are mixed by crossing over with the genes coming from HCI leading to new, efficacious and safe therapeutics for patients.
Giovanni Cincilla is the Chief Scientific Officer at Molomics. Molomics advances the search for structurally novel small molecule therapeutics by Artificial Intelligence (AI) through integrating it with Human Collective Intelligence (HCI).
Leave a Reply