Dr Zhixing Feng is currently a tenure-track assistant professor at Xinhua Hospital affiliated to Shanghai Jiao Tong University School of Medicine. The lab focuses on developing novel computational methods for long-read sequencing data to accelerate biological discoveries and clinical diagnosis. The research areas include: 1. Detecting multiple types of DNA modifications from third-generation sequencing data. Cellular heterogeneity of genomes and epigenomes is prevalent in many biological systems including tumors, microbiome, and pathogens. Although the raw signals of third-generation sequencing data are affected by multiple types of DNA modifications, due to noise and systemic bias, it is computationally challenging to detect DNA modifications directly from third-generation sequencing data. The key informatics problems are high-dimensional data denoising, missing value imputation, dimension reduction, classification, and feature extraction. I developed the first computational models with machine learning and empirical Bayes mixture model to detect multiple types of DNA modifications, including 5mC, 4mC, and 6mA, at a single-molecule level directly from third-generation sequencing data (PLoS CB 2013, NAR 2015). I also applied these models to obtain the complete methylome of Streptococcus pneumoniae, a pathogen causing half a million death globally per year, and revealed the association between methylome and bacterial virulence for the first time (PLoS Pathog 2016). The work is also reported by Nature Review Microbiology as Research Highlight (Nat Rev Microbiol 14, 546). 2. Detecting and phasing low-frequency minor SNVs from third-generation sequencing data Detecting and phasing minor SNVs is the key underlining computational problem in distinguishing multiple conspecific strains in microbiome, viral quasispecies, and detecting somatic mutations. The high error rates compromise the advantage of read length to phasing SNVs using third-generation sequencing. I developed the first computational framework, iGDA, to detect and phase low-frequency minor SNVs (Nature Communications 2021) from third-generation sequencing data. I proposed a new concept called maximal conditional substitution rate, to increase signal-to-noise ratio about 100 times by leveraging linkage among real SNVs. In addition, I proposed two novel algorithms, Random Subspace Maximization and Adaptive Nearest Neighbor Clustering, to resolve the combinatorial explosion problem in estimating the maximal conditional substitution rate and the problem of automatically estimating the number of haplotypes. In testing data, iGDA can detect minor SNVs with a frequency of 0.2% from uncorrected third-generation sequencing data with 15% errors. In the testing metagenomics data, iGDA can accurately distinguish conspecific strains with 0.011% differences, and reconstruct their strain-specific genomes. As sequencing costs rapidly drop, the novel algorithms in iGDA will also play an important role in reconstructing complete full-length cancer genomes and epigenomes.