Combining state - Iktos
17840
page-template-default,page,page-id-17840,ajax_updown_fade,page_not_loaded,,qode-title-hidden,qode_grid_1300,qode-content-sidebar-responsive,qode-theme-ver-10.1.1,wpb-js-composer js-comp-ver-5.0.1,vc_responsive

Combining state of the art computational chemistry and machine learning technologies to solve the issues of new compound design

 

Deep learning Generative models

Generative models are new algorithms based on deep learning technology which enable to generate new data points. After having learned a task (create a picture, write music, generate new content…) the algorithm becomes capable of imagining new images, pieces of music or texts.
Based on this approach and using publicly available databases as an input, Iktos algorithm has learned to design new molecules.

 

Objective functions

To guide the molecular generator in the infinity of the chemical space, we have introduced fitness functions that the algorithm must optimize. Those functions are predictors, specifically built from your project’s data set: your molecules, measured on your assays, activity, selectivity, ADMET… On top of that we also have implemented fitness functions to maximize druggability, synthetic access, and similarity to your initial data set. The molecules we generate are optimized simultaneously on all your success criteria.

 

Top notch predictive models

Iktos molecule generation technology is only as good as the predictors it uses to travel in the chemical space. The quality of the predictors is key.
We therefore have developed ways to construct high-performance predictors and to select the optimal ones, automatically and in a few hours, by benchmarking and optimizing both machine learning algorithms and molecules descriptors (1D, 2D and 3D representations of the molecules, and even data driven representations like latent space for instance).