Identifying the missing proteins in human proteome by biological language model.

BACKGROUND: With the rapid development of high-throughput sequencing technology, the proteomics research becomes a trendy field in the post genomics era. It is necessary to identify all the native-encoding protein sequences for further function and pathway analysis. Toward that end, the Human Proteome Organization lunched the Human Protein Project in 2011. However many proteins are hard to be detected by experiment methods, which becomes one of the bottleneck in Human Proteome Project. In consideration of the complicatedness of detecting these missing proteins by using wet-experiment approach, here we use bioinformatics method to pre-filter the missing proteins.

RESULTS: Since there are analogy between the biological sequences and natural language, the n-gram models from Natural Language Processing field has been used to filter the missing proteins. The dataset used in this study contains 616 missing proteins from the "uncertain" category of the neXtProt database. There are 102 proteins deduced by the n-gram model, which have high probability to be native human proteins. We perform a detail analysis on the predicted structure and function of these missing proteins and also compare the high probability proteins with other mass spectrum datasets. The evaluation shows that the results reported here are in good agreement with those obtained by other well-established databases.

CONCLUSION: The analysis shows that 102 proteins may be native gene-coding proteins and some of the missing proteins are membrane or natively disordered proteins which are hard to be detected by experiment methods.