Benchmarking DNA barcode decoding strategies under high error rates

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• DNA barcoding is crucial for multiplexed biomolecule identification in pooled sequencing.
• High-density barcode arrays for spatial transcriptomics face significant synthesis errors (10-20% per nucleotide).
• Existing error-correcting codes are inadequate for large library sizes and high error rates.

Purpose of the Study:

• To benchmark computational barcode decoding approaches for spatial transcriptomics.
• To evaluate performance under realistic synthesis error conditions and varying barcode lengths.
• To identify the most scalable and accurate decoding method.

Main Methods:

• Benchmarking of three decoding algorithms: Columba, QUIK, and RandomBarcodes.
• Utilized simulated datasets with varying barcode lengths (28-36 nt), library sizes (21k-85k), and error rates (9-32%).
• Validated performance on empirical data from photolithographically synthesized arrays.

Main Results:

• QUIK demonstrated superior recall (87-89%) and precision at medium error rates (~23%) compared to RandomBarcodes and Columba.
• QUIK exhibited significantly higher scalability, processing reads much faster per GPU.
• 34-nt barcodes with QUIK achieved 75% recall at 99.97% precision, outperforming shorter barcodes.

Conclusions:

• QUIK offers the best balance of speed, accuracy, and scalability for spatial transcriptomics with high-density arrays.
• Recommended barcode lengths of 34 nt for achieving high read recovery and precision.
• QUIK is optimized for realistic synthesis error conditions in high-throughput applications.