Faster Antibody Drug Discovery with AI
Antiverse is building a world-first computational antibody drug discovery platform to predict antibody-antigen binding and provide antibody drug candidates.
A combination of state of the art machine learning and phage display techniques is used to predict antibodies that bind to a given antigen target with high affinity. The resulting software can then take antigen target sequence, provided by the customer, and do a high-throughput screening of all possibilities of antibody sequences to detect the sequence that will produce a high-affinity antibody for the target. The customer will be provided with the sequences promptly, thus reducing the time typically required for antibody therapeutics discovery by 3 to 18 months.
Innovation: Combining deep learning with in-house data generation
We are at the cutting edge, leveraging recent advances in deep learning that allow prediction of antibody-antigen binding based solely on the primary structure of the proteins, and harnessing developments in cell-free protein synthesis technology that enable the high-throughput generation of a large, customised proprietary data set to train our bespoke algorithm.
What has hindered the ability to model antibody-antigen interactions so far is both the lack of a suitable data set and the inherent limitations of traditional modelling methods.
The first recent advance is in machine learning. Existing modelling methods typically rely on human expertise to decide which features are relevant, whereas deep learning techniques are successful in great part because they learn directly from the entire dataset. Efforts to predict binding were previously based on 3D structures but the largest publicly available data set, abYsis, only contains around 200 of these, which is far too small for modelling. We are developing and applying recent advances in machine learning (primarily deep learning) to be able to use the amino acid sequence of the antibody and antigen as the data input.
The second recent shift is the improvement in molecular engineering methods. This enables us to generate a large in-house data set of antibody-antigen binding values, coupled to sequence data, with a high-throughput method. A large data set is required for machine learning, but companies offering traditional methods of antibody discovery have very limited data sets because they do not require information about the primary amino acid sequences or 3D structures involved. Thus, the combination of these two breakthroughs opens up the potential to effectively predict antibody sequences for a target of interest. We have close contact with customers who are keen to trial our service, and we have completed a proof of concept on small dataset to understand the technical challenges.
The team consists of 3 co-founders Ben Holland, Dr Rowina Westermeier, Murat Tunaboylu, and a machine learning engineer Ilai Waimann. The project has gradually drawn in collaborators, advisors, potential clients and investment.
Ben has an MEng in Engineering Science from the University of Oxford and experience in mathematical modelling, especially neural networks, and uses the generated data to train the machine learning system.
Murat has a BSc in Electrical Engineering and twelve years of experience as a software engineer. He also has extensive experience in labware automation and is automating the lab work and creating the software that customers will use.
Ilai holds a BSc Biomedical Engineering degree from Israel Institute of Technology and has both machine learning and biomedical lab skills. As part of the machine learning team, he is developing antibody generator models.
Sumit is an expert in assay development and high-throughput compound screening, complex disease gene discovery, drug target discovery in Type II diabetes, obesity and related metabolic disorders, next generation sequencing and analysis of the bacterial biome. He worked at TwistDX, EBI, AstraZeneca, GSK. He holds aDPhil in Genetics from Oxford University, Postdoc positions in Cambridge University, MSc Computer Science from Anglo Ruskin University.
Our academic collaborator and advisor, Andrew Martin, Professor of Bioinformatics and Computational Biology at UCL brings nearly 30 years of antibody research to the table. He is the creator of abYsis, the largest publicly available database for antibody sequences and 3D structures and has led the development and commercialisation of an automated programme for modelling antibodies.
JOIN OUR TEAM
Antiverse is accelerating antibody drug discovery with AI. Join our talented ambitious team to be a part of this revolution!
MACHINE LEARNING ENGINEER
Come and join us in our journey to change drug discovery! We are seeking for a machine learning engineer with a focus on structural biology.
Send us your CV to have a chat: firstname.lastname@example.org
We’ve been selected as one of Hello Tomorrow’s Top 500 deeptech startups from over 4,500 applications from 119 countries!
It is our pleasure to share that Antiverse has closed its Seed Round with well-known, highly regarded investors from software and biotech industries. The capital and the new connections acquired at this round will give Antiverse the chance to bring its machine learning model to the next level.
Delighted to announce Antiverse has won the Accelerate@Babraham 1st Start-up Competition! Looking forward to using the lab space at Babraham for 3 months, with 1-2-1 mentoring from leading experts in science, tech and business, and using the £20,000 cash prize to push our technology forward.
PODCAST - LEARNING WITH LOWELL
40: Computational Antibody Drug Discovery Platform with Machine Learning Startup, Antiverse, with the Co-Founder Murat Tunaboylu
INNOVATE UK GRANT
We are pleased to inform you that our application to Innovate UK
January 2018 sector competition: strand 1, health and life sciences, with
Computational Antibody Design with Machine Learning project has been successful.