AlignBio Tournament 2024: Exazyme Achieves Top-Tier Rankings, Highlighting the Power of Its Cutting-Edge Algorithms

Exazyme proudly participated in the inaugural AlignBio Tournament on May 1, 2023, alongside over 30 distinguished teams from industry and academia. This highly competitive event aimed to benchmark computational capabilities for predicting protein properties and designing new sequences. Exazyme’s powerful algorithms secured first place in the in silico supervised track and fourth place in the in vitro track, showcasing our significant achievements and technological expertise in the field of protein engineering.

The AlignBio Tournament: A benchmarking initiative aiming to accelerate innovation in the computational protein engineering field

The main challenge in protein engineering is developing computational models to analyze and generate protein sequences for various functions. Progress is limited by the lack of benchmarks, comprehensive datasets, and access to experimental characterization. The AlignBio Protein Engineering Tournament tackles these issues by providing a framework for benchmarking different computational methods. Over 30 teams participated across academia and industry, including established market players and Nobel Laureate labs, which highlights the impressive quality of the participants. The tournament featured two rounds: an in silico round for predicting biophysical properties of protein sequences and an in vitro round, where new sequences were designed and experimentally measured.

The theme focused on Enzyme Design, chosen for its broad applications and significance in protein engineering, given its versatility across pharmaceuticals, food processing, biofuels, and environmental remediation. Six datasets were donated by academia and industry,  and lab measurements were performed in-house by a corporate partner. These contributions were crucial for the benchmarking event’s success – thank you! All protocols and data are open-sourced post-tournament to support ongoing advancements in the field.

In Silico Round: Exazyme ranks 1st place in Predicting Protein Properties 

In the in silico round, teams developed models to predict enzyme properties such as expression, thermostability, and specific activity. The round featured two tracks — zero-shot and supervised—with teams choosing their participation. Exazyme entered the supervised track, as our zero-shot modules were still in early development. Teams used training data with amino acid sequences and measured properties to train models and predict properties for new sequences. Events included alkaline phosphatase PafA, α-amylase, β-glucosidase B, and imine reductase. Performance was assessed by comparing predictions to ground-truth values, with top performers advancing to the in vitro round.

The primary metric for evaluation was Spearman’s rank correlation, which is particularly well-suited when benchmarking predictions that will be used for suggesting new sequences. Exazyme ranked among the top three in all four enzyme challenges, achieving first place in the most challenging event, β-glucosidase B. We also ranked first for predicting properties of imine reductase, an enzyme that is of particular importance in pharma production. We secured second place in α-amylase and third place in alkaline phosphatase PafA, ultimately emerging as the leader of the supervised track.

In Vitro Round: Exazyme ranks the 4th place in Designing Protein Sequences, despite very little time and early product stage

The in vitro round evaluated teams on their ability to design protein sequences with specific biophysical properties. Only the top teams from the in silico round advanced to this stage. Exazyme competed alongside other leading teams, designing sequences that were synthesized and experimentally characterized by the Tournament and its partners. Teams received a dataset with single and double mutations from the alpha-amylase enzyme and were tasked with submitting up to 200 amino acid sequences to maximize enzyme activity while maintaining at least 90% of the parent sequence’s stability and expression.

Performance was assessed through a weighted evaluation of protein properties, with submissions ranked against predefined thresholds. Teams showed varied strengths, with SergiR1996 and MediumBio demonstrating superior stability, TUM Rostlab excelling in specific activity, and others facing challenges in synthesis and expression. TUM Rostlab won with the highest-scoring variants and median performance, followed by MediumBio and Marks Lab. Despite being in the early stages of testing a new proposal algorithm, Exazyme ranked 4th overall.

Since the tournament, we have focused on benchmarking and significantly improving our proposal algorithms over the past 13 months. We are excited to participate in the next tournament and demonstrate the advancements we’ve achieved! We are continuously enhancing our product by benchmarking and incorporating more tools to cover a wider range of use cases, making protein engineering challenges even more streamlined for our customers. To learn more about the tournament, the different challenges and results, as well as the participants and donors, read our case study here!