Many machine learning techniques are used as drug discovery tools with the intent to speed characterization by determining relationships between compound structure and biological function. However, particularly in anticancer drug discovery, these models often make only binary decisions about the biological activity for a narrow scope of drug targets. We present a feed-forward neural network, PECAN (Prediction Engine for the Cytostatic Activity of Natural product-like compounds), that simultaneously classifies the potential antiproliferative activity of compounds against 59 cancer cell lines. It predicts the activity to be one of six categories, indicating not only if activity is present but the degree of activity. Using an independent subset of NCI data as a test set, we show that PECAN can reach 60.1% accuracy in a six-way classification and present further evidence that it classifies based on useful structural features of compounds using a within-one measure that reaches 93.0% accuracy.

The face-inversion effect, or the drastic decrease in accuracyseen when a participant is asked to identify inverted faces whencompared to upright faces, is an effect that is not found in objectinversion. Here we suggest a new explanation of this effect usingcomputational models to show that the phenomenon can beexplained by the anatomical mapping from the visual field toprimary visual cortex. We propose that the way inverted faces aremapped onto the cortex is fundamentally different from the wayupright faces are mapped. Our work first shows the advantages ofthis mapping due to its scale and rotation invariance when used asinput to a convolutional neural network. We train the network toperform recognition tasks and show it exhibits scale andrealistically constrained rotation invariance. We then confirm thatthe decline in accuracy seen when a participant is asked to identifyinverted faces is not seen in the network with inverted objectrecognition tasks. With the support of these two findings, we testthe face-inversion effect on our network and are able to show theunique decline in accuracy, suggesting that the way the visual fieldis mapped onto the primary visual cortex is a key facet in themanifestation of this effect.