The number of molecules that can be made by combining atoms of different elements is unfathomably large and it is impossible to make and test all these molecules when searching for new pharmaceuticals or other functional molecules. In fact, the possible molecules are even too many to be tested virtually, by strictly computational methods. However, these computational methods are still faster than experiments and are indispensable tools in modern drug discovery. Particularly desirable are so-called heuristic methods that, instead of screening through all possibilities, may arrive at candidate drug molecules based on information from relatively few of the possible molecules. However, so far, such methods have often failed to distinguish truly excellent from merely good molecules, and expert chemists typically must spend considerable time to identify the most interesting candidates produced by the heuristic methods. Seeing a pressing need for improvement, we will integrate expert knowledge early on to help steer the heuristic methods all the way to the most interesting candidates, the ones with the greatest possibility to end up as pharmaceuticals. To pursue this new approach, we have assembled an interdisciplinary team of leading researchers in chemistry, biomedicine, and data science. eHACS includes research on new, enabling technologies as well as their deliberate application to accelerate drug discovery.
The initial focus is on validating and parameterizing computational methods for estimating the affinity with which candidate inhibitors bind to their target enzymes. The most accurate and computationally inexpensive methods will subsequently be used to estimate binding affinities of candidate inhibitors in actual drug design.
Computational molecular design is vital for identifying functional molecules to address critical societal challenges, such as the accelerating antibiotic resistance and our demanding combat of cancer. This quest, however, faces a combinatorial explosion in the number of potential drug candidates - the chemical space made up of all small drug-like molecules - which is unfathomably large. Testing any significant fraction of these molecules, either computationally or experimentally, is simply impossible. Thus, heuristic optimization methods, or automated 'de novo' design, is the long-standing Holy Grail of drug discovery. Unfortunately, these methods have so far failed to deliver on their promise and must be complemented by inefficient and slow post-processing involving expert chemists. Having advanced the international forefront of automated de novo design, we see a pressing need for an alternative approach. We anticipate that a new, knowledge-guided molecular design, integrating the expert early on, will help focus and steer the overall optimization and drastically reduce the need for manual post-processing. To unlock these possibilities, knowledge-based expert guidance, modern molecular design, and empowering machine learning must be integrated. To research and pursue this new approach, we have assembled an interdisciplinary team of leading scientists from chemistry, biomedicine, and data science (visual data science as well as machine learning). The eHACS project includes research on new, enabling technologies (data-science-empowered solutions) as well as their deliberate application to overcome challenges in drug discovery and to design potent inhibitors of targets for cancer and antibiotics.