Simplified, interpretable graph convolutional neural networks for small molecule activity prediction

We here present a streamlined, explainable graph convolutional neural network (gCNN) architecture for small molecule activity prediction. We first conduct a hyperparameter optimization across nearly 800 protein targets that produces a simplified gCNN QSAR architecture, and we observe that such a model can yield performance improvements over both standard gCNN and RF methods on difficult-to-classify test sets. Additionally, we discuss how reductions in convolutional layer dimensions potentially speak to the “anatomical” needs of gCNNs with respect to radial coarse graining of molecular substructure. We augment this simplified architecture with saliency map technology that highlights molecular substructures relevant to activity, and we perform saliency analysis on nearly 100 data-rich protein targets. We show that resultant substructural clusters are useful visualization tools for understanding substructure-activity relationships. We go on to highlight connections between our models’ saliency predictions and observations made in the medicinal chemistry literature, focusing on four case studies of past lead finding and lead optimization campaigns.

Supervised topic modeling for predicting molecular substructure from mass spectrometry

Small-molecule metabolites are principal actors in myriad phenomena across biochemistry and serve as an important source of biomarkers and drug candidates. Given a sample of unknown composition, identifying the metabolites present is difficult given the large number of small molecules both known and yet to be discovered. Even for biofluids such as human blood, building reliable ways of identifying biomarkers is challenging. A workhorse method for characterizing individual molecules in such untargeted metabolomics studies is tandem mass spectrometry (MS/MS). MS/MS spectra provide rich information about chemical composition. However, structural characterization from spectra corresponding to unknown molecules remains a bottleneck in metabolomics. Current methods often rely on matching to pre-existing databases in one form or another. Here we develop a preprocessing scheme and supervised topic modeling approach to identify modular groups of spectrum fragments and neutral losses corresponding to chemical substructures using labeled latent Dirichlet allocation (LLDA) to map spectrum features to known chemical structures. These structures appear in new unknown spectra and can be predicted. We find that LLDA is an interpretable and reliable method for structure prediction from MS/MS spectra. Specifically, the LLDA approach has the following advantages: (a) molecular topics are interpretable; (b) A practitioner can select any set of chemical structure labels relevant to their problem; (c ) LLDA performs well and can exceed the performance of other methods in predicting substructures in novel contexts.

Zooming through scales by the billion

“The climate is what you expect, the weather is what you get”: The climate is a kind of average weather. But is it really? Those of us who have thirty years or more of recall are likely aware of subtle but systematic changes between today’s weather and the weather of their youth. I remember Montreal winters with much more snow and with longer spells of extreme cold. Did it really change? If so, was it only Montreal that changed? Or did all of Quebec change? Or did the whole planet warm up? And which is the real climate? Todays’ experience or that of the past? The key to answering these questions is the notion of scale, both in time (du­ration) and in space (size). Spatial variability is probably easier to grasp because structures of different sizes can be visualized readily (Fig. 1.1). In a puff of cigarette smoke, one can casually observe tiny wisps, whirls, and eddies. Looking out the window, we may see fluffy cumulus clouds with bumps and wiggles kilometers across. With a quick browse on the Internet, we can find satellite images of cloud patterns literally the size of the planet. Such visual inspection confirms that structures exist over a range of 10 billion or so: from 10,000 km down to less than 1 mm. At 0.1 mm, the atmosphere is like molasses; friction takes over and any whirls are quickly smoothed out. But even at this scale, matter is still “smooth.” To discern its granular, molecular nature, we would have to zoom in 1,000 times more to reach submicron scales. For weather and climate, the millimetric “dissipation scale” is thus a natural place to stop zooming, and the fact that it is still much larger than molecular scales indicates that, at this scale, we can safely discuss atmos­pheric properties without worrying about its molecular substructure. Clouds are highly complex objects. How should we deal with such apparent chaos? According to Greek mythology, at first there was only chaos; cosmos emerged later.

Tuning the thermal conductivity of methylammonium lead halide by the molecular substructure

It is shown by molecular dynamics that the substructure of organic molecules can tailor the thermal conductivity of MAPI.