Feature Selection for Topological Proximity Prediction of Single-Cell Transcriptomic Profiles in Drosophila Embryo...

Area of Science:

• Developmental Biology
• Computational Biology
• Genomics

Background:

• Single-cell transcriptomics loses spatial information during cell isolation.
• In situ hybridization patterns offer complementary spatial data.
• The DREAM Single Cell Transcriptomics Challenge (SCTC) aimed to predict cell locations using gene expression.

Purpose of the Study:

• To detail a genetic algorithm (GA) for selecting informative genes in single-cell spatial prediction.
• To analyze GA performance and parameterization for improved feature selection.
• To provide insights into gene-set selection for topological association prediction.

Main Methods:

• Application of a genetic algorithm (GA) combined with the DistMap algorithm.
• GA was used to identify genes carrying positional and proximity information.
• Post-challenge parameterization and analysis of the GA method were performed.

Main Results:

• The GA-based gene selection performed well, ranking in the top 10 for two SCTC sub-challenges.
• Detailed implementation and parameterization of the GA are discussed.
• The study identifies areas for improvement in GA-based gene-set selection.

Conclusions:

• GA is a viable approach for selecting informative genes in single-cell spatial prediction.
• This work offers valuable insights into feature-selection strategies for single-cell analysis.
• The findings complement existing consortium research on spatial transcriptomics.