DNAMIC CODEC Poster for MoleculArXiv Workshop 2026

Photolithographic synthesis, for instance, provides greater scalability for producing volumes in the range of gigabyte and beyond, potentially at lower cost (Gimpel et al., 2024). However, this approach poses challenges reminiscent of those encountered in earlier methods. Notably, the synthesized oligonucleotides are smaller and have a much higher error rate than those available from providers using silicon-based synthesis (Antkowiak et al., 2020).

Homogeneity is also a concern, as variations in error rates depending on the position of the synthesis on the photolithographic chip can lead to uneven prevalence of certain strands. Contrary to the trend of developing codecs capable of easier and faster decoding thanks to low-error synthesis, we need to find algorithms that are more resilient to synthesis (and sequencing) errors. Our solution consists of proceeding according to a multistep correction system: the encoding scheme divides the files into separate logical blocks, allowing us to decode them independently with the appropriate level of precision.

For the clustering stage, we use a custom algorithm based on k-mers and a state-of-the-art approximate nearest-neighbor method, which we implemented with Nvidia's cuVS for greater efficiency. To allow comparisons between silicon-based synthesis and photolithographic synthesis, the results on data showing different error profiles will be presented.