RECONSTRUCTING DNA COPY NUMBER BY PENALIZED ESTIMATION AND IMPUTATION.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• DNA Copy Number Variation (CNV) is ubiquitous and detectable with modern genotyping platforms.
• Hidden Markov Models (HMMs) are commonly used for CNV data interpretation.
• Existing methods for CNV reconstruction face computational challenges.

Purpose of the Study:

• To develop a more effective and computationally efficient method for CNV reconstruction.
• To improve upon the fused-lasso penalty approach for CNV estimation.
• To explore an alternative discrete optimization framework for CNV imputation.

Main Methods:

• Modified fused-lasso penalty with a smooth approximation to the absolute value function.
• Developed a novel Majorization-Minimization (MM) algorithm for optimization.
• Applied a fast Newton's method for joint parameter updates.
• Reframed CNV reconstruction as discrete optimization and imputation using dynamic programming.

Main Results:

• Successfully minimized the fused-lasso criterion effectively.
• Achieved accurate CNV imputation comparable to HMMs.
• Demonstrated substantially lower computational cost than existing methods.
• The discrete optimization approach leverages information ignored by current fused-lasso methods.

Conclusions:

• The proposed methods offer a highly effective and computationally efficient solution for CNV reconstruction.
• Discrete optimization provides an accurate and faster alternative for CNV imputation.
• This work advances the analysis of genomic CNV data.