On March 3, 2023, Dr. Jianhua YAO, Chief Scientist of AI Healthcare at Tencent AI Lab was invited to the 102^nd Science Lecture in College of Science. He gave a lecture themed “Researches and Applications of Artificial Intelligence in Genomic Computing”, which was chaired by Prof. Ruijun TIAN of the Department of Chemistry, SUSTech.

With the development of omics technologies, computational biology has become the key factors for biology.

In the lecture, Dr. YAO first reviewed the composition of the central principle and the development history of sequencing technology, and pointed out that with the breakthrough of single-cell sequencing technology in recent years, conditions have been created for answering questions in life science. Next, he listed the problems and challenges encountered in the current field, including: high data dimension, disordered gene expression data, no existing methods to make full use of prior knowledge of interaction between genes, widespread data missing, batch effect, etc. Dr. YAO suggests that an efficient, universal and unified model of genetic coding based on deep learning could help solve this problem.

Dr. YAO introduced scBERT: A single-cell annotated model based on a large-scale pre-training language model that effectively transforms expression profile data as model inputs by targeting gene coding and expression coding, reduces data noise, and provides high-resolution genome-wide level feature interpretation without dimension reduction and screening for feature genes by using a Performer encoder. He introduced the research results based on single-cell proteome coding characteristics and spatial omics techniques.

With the rapidly advance in gene sequence methods, novel AI algorithm for analysis of complex big data is needed. Dr. YAO said that AI plays an essential role in gene-regulated network of multi-omics, spatial omics and so on, providing tools for life scientists. And then he highlighted the AI model-based sequencing and analysis of immune libraries, which can greatly reduce researchers' efforts to screen antigen-antibody binding epitopes.

In the interactive question-and-answer session, the audience actively asked questions about scBERT, the amount of data required for the model, and the applicable scenarios of the model. Dr. YAO gave detailed answers.

Q: What magnitude of data is required to design the model?

A: The larger the amount of data, the higher the robustness of the model. Based on the verification of wet experiment, the accuracy and sensitivity of the model are improved by iteration.

Q: How to choose the right algorithm?

A: Based on the understanding of the field, select mature and classic tools for verification one by one. Select appropriate tools or make personalized adjustments according to a common benchmark.

In conclusion, Prof. TIAN handed an honorary certificate to Dr. YAO and had a photo with him.