Overview

The discovery/creation of novel proteins with desired functional properties is one of the fascinating challenges in biology and demonstrates a comprehensive understanding of proteins’ interaction mechanisms. The discovery of novel proteins can open the way to practical applications in a plethora of fields.

At BioMetis, we use the integration of sequence-based, structure/dynamics-based, and function-based methods to discover desired novel proteins. Next-generation sequencing techniques, the continuous increase in the number of protein structures, and advanced Molecular Dynamics simulation techniques provide a sound basis for novel protein discovery. The function-based approach includes conventional Quantum mechanics calculation for function design and emerging Artificial Intelligence approaches to transform the Big Data into rules for discovering novel functional proteins. 

We have developed a very comprehensive sequence/text-based patent search and analysis pipeline. Any leading candidates from the design space will be carried on for straight-forward FTO and prior art analyses. 

Protein Engineering and strain optimization are difficult tasks, owing in part to the highly rugged fitness landscape of proteins and genome and the underlying epistatic interactions across different amino acids in one or more proteins. We could adventure many routes and still fail to discover the best route leading to the global optimum, a magic sequence with desirable blend of biochemical and biophysical properties.

Artificial Intelligence has recently demonstrated its power in the fields such as image processing and speech recognition, but breakthroughs in the protein engineering and strain improvement field are much limited.

At BioMetis, we develop machine learning and deep learning frameworks to achieve a comprehensive understanding of complicated epistatic interactions between amino acids. We use a small set of high-quality experimental data to generate in silico data, leading to more predictive models. Our AIDME technology significantly reduces the rounds of iteration in directed evolution, showing the optimization course to be quicker, more informative, and cost-effective. 

Using our pioneering mutant synthesis technology and method, unlike a traditional one, allows the analysis of many diverse mutations and specific mutation combinations with tunable mutation rates. These proteins and strain variants contain characteristics critical for AI system to learn the data more effectively. Consequently, our high-quality collections lead to more fruitful hits and more effective learning by AI. 

Errors and variations in data are one of the significant challenges in AI technology development for protein and genome engineering. The high-quality datasets generated by robotic automation and stored in our proprietary database provides a solid support for the AIDME platform.