The never-ending emergence of SARS-CoV-2 variations of concern (VOCs) has challenged the whole world for pandemic control. In order to develop effective drugs and vaccines, one needs to efficiently simulate SARS-CoV-2 spike receptor-binding domain (RBD) mutations and identify high-risk variants. We pretrain a large protein language model with approximately 408 million protein sequences and construct a high-throughput screening for the prediction of binding affinity and antibody escape. As the first work on SARS-CoV-2 RBD mutation simulation, we successfully identify mutations in the RBD regions of 5 VOCs and can screen millions of potential variants in seconds. Our workflow scales to 4096 NPUs with 96.5% scalability and 493.9× speedup in mixed-precision computing, while achieving a peak performance of 366.8 PFLOPS (reaching 34.9% theoretical peak) on Pengcheng Cloudbrain-II. Our method paves the way for simulating coronavirus evolution in order to prepare for a future pandemic that will inevitably take place. Our models are released at https://github.com/ZhiweiNiepku/SARS-CoV-2_mutation_simulation to facilitate future related work.

DNA methylation data-based precision tumour early diagnostics is emerging as the state of the art for molecular tumour recognition, which could capture the signals of cancer occurrence 3 ~ 5 years in advance and clinically more homogenous groups. However, the sensitive of early detection for many tumors is about 30%, which needs to be greatly improved. Nevertheless, on the basis of the whole genome bisulfite sequencing methylation data, a comprehensive characterisation of the entire molecular genetic landscape of the tumor as well as the subtle differences between different tumours could be identified. With the accumulation of methylation data, high performance deep learning models that considering and modeling more unbiased information need to be developed. According to the above analysis, we designed a pipeline to investigate genome-wide DNA methylation patterns for precision multi-tumour early diagnostics. We proposed a graph convolutional network considering the attention mechanism to dissect DNA methylation heterogeneity of different cancer types. The attention mechanism in the graph convolutional network architecture could detect the cancer signals from genome-wide methylated molecular interactions, improving the robustness and sensitivity of multi-tumor classification. Experimental results demonstrates the feasibility of precision multi-tumour early diagnostics with the DNA methylation data. The workflow presented here is very useful for tumor classification, which highly relevant for the future blood diagnosis and treatment of the tumour.

DNA methylation data-based precision tumor early diagnostics is emerging as state of the art technology, which could capture the signals of cancer occurrence 3∼5 years in advance and clinically more homogenous groups. At present, the sensitivity of early detection for many tumors is about 30%, which needs to be significantly improved. Nevertheless, based on the genome wide DNA methylation information, one could comprehensively characterize the entire molecular genetic landscape of the tumors and subtle differences among various tumors. With the accumulation of DNA methylation data, we need to develop high-performance methods that can model and consider more unbiased information. According to the above analysis, we have designed a self-attention graph convolutional network to automatically learn key methylation sites in a data-driven way for precision multi-tumor early diagnostics. Based on the selected methylation sites, we further trained a multi-class classification support vector machine. Large amount experiments have been conducted to investigate the performance of the computational pipeline. Experimental results demonstrated the effectiveness of the selected key methylation sites which are highly relevant for blood diagnosis.