Log-Sum Heuristic Recovery for Automated Isoform Discovery and Abundance Estimation from RNA-Seq data

The recent RNA-Seq technology brings computational challenges in transcriptome assembly and analysis. Lasso-type methods were designed to address the ambiguity and unidentifiability issues in isoform discovery and abundance estimation from RNA-Seq data. However, typical Lasso-type methods are confined to taking the l₁ norm to approximate the desired l₀ norm, but such approximation has been shown to be limited in analytical performance and modeling of the specific computational problem. The isoform discovery and quantification tasks still face the challenge of high-level false positive/negative predictions. In this paper, we propose the Log-Sum Heuristic Recovery for Automated Isoform Discovery and Abundance Estimation method, which is attempted at a closer approximation to the l₀ norm and more effective modeling of the parsimony principle involved in isoform discovery. The method is applied to transcriptome analysis with RNA-Seq data. Both simulation and real data experiments demonstrate that the proposed method is promising to be an effective computational tool for isoform discovery and quantification.