With the rapid applications of high throughput technologies such as next generation sequencing, it is particularly urgent to detecting biomarkers arising from different high-throughput platforms. In this talk, I will elaborate recent research effort in my group that aims to integrate the omics profiles for cancer diagnosis and prognosis prediction. We have constructed a versatile individual-oriented pathway-based modeling framework from multiple omics data types to predict patient prognosis and/or diagnosis. This approach performs better than those models built on the end-level measurables (genes, metabolites, SNPs etc) and they are relatively easy to transfer between data types and integrate among data types. More recently, we have employed deep-learning method for multi-omics based prognosis biomarker investigation, and demonstrated the robust predictive power in multiple liver cancer population cohorts. Lastly, I will report our most recent work on heterogeneity detection among tumor single-cell RNA-Seq data, using Small Nucleotide Variation (SNVs) as the new, alternative molecular features.