2 of 17

Intros

Mark Pitman CBDO

BizDev in Biotech/Pharma

Nikolai Russkikh CEO

ML/AI research engineer in Biotech

3 of 17

Problem

Balancing Complex Property Trade-offs in a Vast Design Space

4 of 17

There are 5.29e+13 of variants to impose just 4 substitutions on 300 aa protein

Vastness

5 of 17

Beneficial mutations are rare.

Scarcity

6 of 17

Enhancing one property often risks degrading others, threatening the industrial viability of the enzyme.

Risks of losing industrially critical properties

7 of 17

Solution

III

Simulated fitness landscape

8 of 17

Scan candidates in-silico while assessing uncertainty

Simulated fitness landscape

AI regression model

9 of 17

Do 2x less wet-lab experiments while reaching 10000x more variants

Simulated fitness landscape

Set up predictive model with this

Universe of variants

Accessible in silico

Accessible in wet lab

10 of 17

NB and Enzymes/Proteins

Neoncorte Bio’s AI tools applied to enzymes and proteins

11 of 17

The Process

We provide a UI to regression model for testing individual hypotheses and to navigation a large database of candidates assessed in cilico to support the decision making

12 of 17

The Process

Initial training set composition

Literature data

Existing customer data

Finding mutation tolerant sites with Protein Language Models

Heuristics (polarity, size etc)

13 of 17

NB Optimization Flow

Proposing candidates with good high predicted fitness

Wet lab assessment

Incorporating assessed variants into the dataset

Training a better sequence-to-activity model

15 of 17

Team

With broad experience in AI applications and software engineering within the life sciences, we excel at understanding and meeting our customers' unique needs. Some of our key projects include:

Automated NGS Data Analysis Platform: Developed a production-grade solution for the automated processing, annotation, and analysis of Next-Generation Sequencing (NGS) data.
Award-Winning Single-Cell Data Integration: Created state-of-the-art solutions for diagonal integration of multimodal single-cell data, recognized with a NeurIPS prize.
Metagenomic Taxonomic Classification: Designed advanced algorithms for classifying metagenomic sequencing reads.
High-Throughput Base Calling Pipeline: Developed an efficient pipeline capable of processing millions of sequencing images.
Cell Counting via Computer Vision: Implemented a computer vision solution for accurate counting of cells in microphotography images.