Mantis-ML is a framework for the identification of novel gene-disease associations. To dive into the results of running Mantis-ML across the human phenome, simply enter a disease or gene name into the search box.

Overview

---

Mantis-ML works by modelling the biological context of disease by leveraging (i) annotation terms: these are signalling pathways, biological processes, tissues etc., known to be implicated in the disease of interest, and (ii) an input gene list: containing a set of genes as positive examples, these are genes confirmed to be associated to the disease. Mantis-ML models the broader biological context underlying the input gene list, identifies the features they share in common, and eventually uncovers novel, under-reported genes, which are likely to be associated to the disease in a similar way to the input genes. Additional details of the methodology are provided in the About page.

A schematic of the methodology is shown above. In effect, a disease is chosen by the user from one of 1000s present across multiple resources. After having selected a disease, natural language processing techniques are leveraged to identify features that are relevant to that disease. Having isolated these features, Mantis-ML is trained to predict novel disease-gene associations based on genes that are known to be associated to a given disease. Phenome-wide Mantis-ML presents a fully streamlined, automated and scalable workflow for the compilation and deployment of Mantis-ML to hundreds of diseases/phenotypes; subsequent aggregation of the results from each individual run, and calculation of summary statistics.

Summary of the results

---

For each disease, Mantis-ML reports the following