Function and Feedback on Our Project
As an enzyme with RNA endonuclease activity, the ability of KmAgo to perform its activity at physiological temperatures determines its feasibility as a feasible RNA tool. In the targeted modification of KmAgo, the trial and error of mutations at different
sites consumes enormous resources. Therefore, we need to quickly screen feasible mutation sites through non experimental methods.
This project aims to develop a feasible and high-precision prediction software, based on deep learning methods, to directly predict the thermal stability (melting temperature, Tms) of proteins through protein sequences, in
order to achieve rapid screening of feasible KmAgo mutation sites.
Through our software, users can predict the thermal stability of target proteins (using KmAgo as an example) and use this as a standard to filter out mutation sites with poor thermal stability, thus achieving low-cost directed
mutations. This helps with directed mutations of KmAgo, the design of specific functional enzymes, and industrial production.
Considering the current demand for rapid and low-cost screening of feasible KmAgo mutation sites, we have constructed a complete protein thermal stability prediction model based on sequence information and provided an online software based on this.
Our model is trained for KmAgo, analyzed for protein structure based on ESM2, and regressed downstream. The modular segmentation of the model itself also provides convenience for future algorithm optimization and the introduction
of more relevant functional modules.
Our model provides a feasible solution for predicting the thermal stability of KmAgo's fixed-point mutants based on sequence information. The model can be used to analyze the structure of KmAgo and predict the melting temperature
of the corresponding structure. By excluding a large number of low heat stable mutants, users can quickly screen out the most feasible mutation sites. At the same time, online software simplifies the process of using the model for
experimental personnel, reducing experimental costs and time costs.
The emergence of CRISPER-CAS9 technology provides extremely convenient tools for DNA editing, and for RNA editing, current researchers consider KmAgo as a feasible solution with broad prospects. In KmAgo directed mutations
targeting RNA endonuclease activity under physiological temperature conditions, researchers can batch input sequence information of KmAgo mutants after site-specific mutations and use our model to predict the melting temperature of
these mutants. Using melting temperature as an indicator to determine the structural stability of mutants, researchers can conduct experimental verification on the mutant with the best thermal stability, and further predict and mutate
in the next round, ultimately obtaining the directed evolution of KmAgo. In addition, our model can be applied in the industrial production of enzymes, reducing the number of iterations required to produce industrial enzyme products.
Due to budget constraints, the capacity of the servers we have rented is limited. Therefore, we have initially uploaded our first version of the model for the development and trial of the web-based software.
We have provided the web-based software for testing to the Tsinghua team and senior students in our collaborating lab. On one hand, they have pointed out that the current software's accuracy is not yet optimal. Despite performing
well in tasks related to mutation effect classification and thermal stability ranking, there are still noticeable errors when predicting specific TM values. This has become a primary focus for us in the subsequent software iterations
and improvements. On the other hand, they have suggested that the current feedback mechanism is not very user-friendly, and it could be improved by sending results to users via email. We will address this in the upcoming server maintenance
as well.