FusionPro: A Recurrent -Neural-Network-based Program to Accelerate Construction of Fusion Proteins towards Drug Discovery

Overview

School of Medicine Fusion proteins play a critical role in drug discovery by modeling protein complexes, facilitating protein purification, and allowing researchers to monitor expression. Prerequisite to visualize their structure is the creation of clonable DNA constructs th at encode the desired proteins. There are numerous processes to create these recombinant plasmids. IVA (In Vivo Assembly) cloning has emerged as the most efficient method; however, planning for fusion protein assembly using IVA is error -prone and time-consuming. To address these issues, FusionPro utilizes recurrent neural networks and validation programs to generate optimal primer sets and a corresponding, clonable DNA construct. The program eliminates the production of a sequence that encodes the wrong pro tein by generating all possible open reading frames from the inputted plasmid sequences and leveraging them to verify the final construct. It optimizes the linker and primers through a bidirectional long short -term memory -based architecture that uses the p lasmid sequences as context. FusionPro also significantly reduces planning time from hours to under fifteen minutes by cutting the myriad of components researchers must consider to seven user inputs. The program was tested to create four versions of a GluA 1-𝛾5 tethered construct; western blot analysis confirmed the successful creation of all versions of the fusion protein. DNA sequencing further showed that the recombinant plasmids created using the generated primers matched the constructs predicted by FusionPro. This research illustrates that leveraging artificial intelligence in fusion protein assembly may yield optimal results while significantly reducing planning time, therefore accelerating protein structure research and drug discovery.