Search

Molecular design benefits from a strong partnership between chemical intuition and machine learning. Given the proliferation of machine learning in the small molecule catalyst space, several ideas ...

Given a particular descriptor/method combination, some quantitative structure-activity relationship (QSAR) datasets are very predictive by random-split cross-validation while others are not. Recent literature in modelability suggests that the limiting issue for predictivity is in the data, not the QSAR methodology, and the limits are due to activity cliffs. Here, we investigate, on in-house data, the relative usefulness of experimental error, distribution of the activities, and activity cliff metrics in determining how predictive a dataset is likely to be. We include unmodified in-house datasets, datasets that should be perfectly predictive based only on the chemical structure, datasets where the distribution of activities is manipulated, and datasets that include a known amount of added noise. We find that activity cliff metrics determine predictivity better than the other metrics we investigated, whatever the type of dataset, consistent with the modelability literature. However, such metrics cannot distinguish real activity cliffs due to large uncertainties in the activities. We also show that a number of modern QSAR methods, and some alternative descriptors, are equally bad at predicting the activities of compounds on activity cliffs, consistent with the assumptions behind "modelability." Finally, we relate time-split predictivity with random-split predictivity and show that different coverages of chemical space are at least as important as uncertainty in activity and/or activity cliffs in limiting predictivity.

An important metric for gauging the impact a synthetic route has on chemical resources, cost, and sustainability is process mass intensity (PMI). Calculating the overall PMI or step PMI for a given synthesis from a process description is more and more common across the pharmaceutical industry, especially in process chemistry departments. Our company has established a strong track record of delivering on our Corporate Sustainability goals, being recognized with eight EPA Green Chemistry Challenge Awards in the last 15 years and we show how these routes help define aspirational PMI tar-gets. While green chemistry principles help in optimizing PMI and developing more sustainable processes, a key challenge for the field is defining what a ‘good’ PMI for a molecule looks like given its structure alone. An existing tool chemists have at their disposal to predict PMI requires the synthetic route be provided or proposed (e.g., via retrosynthetic analysis) which then enables practitioners to compare predicted PMIs between routes. We have developed SMART-PMI (in-Silico MSD Aspirational Research Tool) to fill the gap in predicting PMI from molecular structure alone. Using only a 2D chemical structure, we can generate a predicted SMART-PMI from a measure of molecular complexity. We show how these predictions correlate with historical PMI data from our company’s clinical and commercial portfolio of processes. From this SMART-PMI prediction, we have established target ranges which we termed “Successful”, “World Class”, and “Aspirational” PMI. The goal of this range is to set the floor for what is a “good” PMI for a given molecule and provide ambitious targets to drive innovative green chemistry. Using this model, chemists can develop synthetic strategies that make the biggest impact on PMI. As innovation in chemistry and processes lead to better and better PMIs , in turn, this data can drive ever more aggressive targets for the model. The potential of SMART-PMI to set industry-wide aspirational PMI tar-gets is discussed.