Alexander Kroll

The paper shows that high accuracy in predicting enzyme kcat fold-changes from sequence data often results from data biases, not true mechanistic understanding, and introduces a bias-aware framework to improve predictions.

FoldVision is a compute-efficient 3D CNN for proteins that outperforms sequence-based models on structure-dependent tasks using atom-level representations.

This paper evaluates the DLKcat model's ability to predict kcat values for enzymes and mutants, finding it unreliable for low sequence identity and poor at capturing mutation effects.

This paper introduces SPOT, a machine learning model that predicts substrates for transport proteins with high accuracy, outperforming existing methods.

ProSmith predicts drug-target interactions using a multi-modal Transformer Network, showing better generalization and performance than previous models.

This paper introduces TurNuP, a model predicting enzyme turnover numbers for natural reactions, outperforming existing methods and generalizing across enzyme similarity.

This paper introduces ESP, a machine learning model predicting enzyme-substrate pairs with over 90% accuracy, outperforming existing models for specific enzyme families.

The paper develops a deep learning model to predict Michaelis constants for enzyme-substrate combinations using structural features.