Ligand-based virtual screening is a widely used application of artificial intelligence for drug discovery. A common approach is to encode the chemical motifs within a set of molecular structures as a binary fingerprint vector, and then use machine learning methods to predict the activity or inactivity in receptor binding screening assays. However, such ligand-based methods are hindered when chemical motifs exist simply by chance and play no role in interacting with the receptor. Recently, it has been observed that artificial neural network architectures that more explicitly mimic the architecture of biological networks can have significant advantages for limited training data. One of these—the Drosophila olfactory network—implements a locality and novelty-sensitive Bloom filter algorithm. Bloom filters are an efficient way of “remembering” if an item has been encountered using only a very small amount of storage space. This variant calculates the similarity of a query item to previously stored inputs. This is relevant to the problem of drug design because it provides a method for evaluating the structural similarity between a small set of molecules known to be chemically active with a receptor compared to an unknown molecule. Using an experimental dataset of candidate drugs for the human muscarinic acetylcholine receptor M1, a target for diseases such as Alzheimer’s and schizophrenia, we compared the performance of a simulated Drosophila olfactory network to other traditional approaches for ligand based-screening.

Applying the <i>Drosophila</i> olfactory network to drug discovery

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.

Applying the Drosophila olfactory network to drug discovery