Our project explores the relationship between biological activities and temperature, approaching the topic from both micro and macro perspectives.
From micro level, enzymes, as essential components of life, have long been a research focal point. One of their fundamental characteristics, optimal temperature, serves as a key link between biological activity and temperature. However, due to the need for numerous repetitive experiments to measure the optimal temperature in wet labs, the currently available data is very scarce. Therefore, the method of predicting with AI has been proposed. Considering the low accuracy of existing models, we used the Hyena framework to build our own high-precision model, HEATMAP-AI.
From macro level, Genome-scale metabolic (GEM) models treat cells holistically and are commonly used in metabolic research. However, neither GEM nor its enhanced model, ecGEM, account for the impact of temperature on cells, largely because the effect of temperature on enzyme activity is difficult to quantify. Therefore, we used our HEATMAP-AI to predict the optimal temperature, quantify enzyme activity at specified temperatures, and constructed an etcGEM for \( Saccharopolyspora\:spinosa \) incorporating temperature constraints.
Existing protein language models struggle with accurate enzyme temperature predictions. To tackle this, our team curated enzyme temperature data and built a deep learning model using the Hyena framework. The resulting HEATMAP-AI model significantly outperforms previous models, and we applied it specifically to the metabolic enzymes of \( Saccharopolyspora\:spinosa \).
For more information, please visit Software section and GitLab repository. To try this model, please visit Website and Implementation section.
While GEM and ecGEM are powerful tools for studying metabolism, they lack temperature considerations. By integrating HEATMAP-AI predictions, we estimated enzyme activities across temperature gradients and constructed an etcGEM model for \( Saccharopolyspora\:spinosa \) with temperature constraints.
For more information, please visit Model section and GitLab repository.
To train HEATMAP-AI for enzyme temperature prediction, we collected and organized data from public databases and scientific literature, which will be a valuable resource for future researchers.
To download the data, please visit Website.
After constructing the deep learning model HEATMAP-AI, to build the etcGEM, our team predicted the optimal temperature of metabolic enzymes in \( Saccharopolyspora\:spinosa \).
For more information, please visit Model section and GitLab repository.
\( Saccharopolyspora\:spinosa \) is widely used in industrial production, but existing GEM models are of limited accuracy. Therefore, to construct the etcGEM model, we first constructed the ecGEM model. Based on this, our team combined our trained deep learning model to construct the etcGEM.