Using a Graph Recommendation Algorithm for Predicting Chemical—Cell Interaction
Recommendation algorithms are often written with a user-product relationship in mind. “Which user bought what product” or “Who liked this movie”. From this data, a prediction is made about an unknown user-product relation.
However, this same algorithm can predict reactions between a chemical and a specific cell line.
In this blog, I will implement a Collaborative filtering algorithm in a graph database. The focus will be on how to implement it. While a basic validation is done, it is not the focus of this blog.
The Data
The dataset used is the NCI60 dataset. In a previous blog, I went into detail on how I created the full graph, here is just a recap.
The NCI60 dataset has the GI50 measurement. This is the concentration of a chemical to have a 50% Growth Inhibition of a cell line.
Concentration is given in the logarithmic scale, which means a GI50 of -5 means the concentration is 10–5 or 0.0001
The Graph
The graph I made previously holds the experiments, their conditions, their measurements, and all variables used. This is done with the idea it can be extended later with different kinds of experiments.