Platform

Machine Learning Modelling

Frobenius selects from a diverse library the most appropriate algorithms to model input data, optimising for data set size and signal to noise ratio.  Each model understands its applicability domain and accurately estimates its error.

Bayesian Statistics

Frobenius predicts a distribution of possible outcomes for each compound across every modelled endpoint using a Bayesian framework. This approach enables the aggregation of models built on data sources of variable quality.

Generative Chemistry

Using bespoke generative algorithms Frobenius populates queried regions of chemical space with a diverse set of synthetically accessible compounds following a range of design rationales.

Compound Selection

Frobenius balances exploration and exploitation of chemical space in the set of compounds it proposes, maximising the likelihood of success within that round of synthesis. 

Data Network

Hecke collates all ligand data associated with a protein creates an interaction network describing the relationships implicit in the data.

Network Validation

Hecke corroborates the target relationships with validation models then isolates those with significant predictive power.

Target Prioritization

Hecke can appraise the prospective druggability of understudied targets, providing a quantitative analysis to facilitate reasoned target prioritization.