How they use Nebius AI Cloud

Helical trained Helix-mRNA on Nebius AI Cloud and uses the infrastructure to adapt biological foundation models to specific datasets. Its virtual research tools help scientists explore DNA, RNA and single-cell data. Its Virtual AI Labs let pharma and biotech teams scale virtual experiments in hours rather than months. Helix-mRNA captures long-range biological signals and can be quickly adapted to proprietary data. The platform supports target discovery, patient stratification and RNA design using single-cell, RNA and DNA foundation models. Nebius clusters, low-latency connectivity and integrated file storage support distributed training and model alignment.

Base pretraining of the 500-million-parameter model used 16 GPUs and took 10 days. Training drew on mRNA sequences from viruses, plants, fungi, humans and other animals. Specialization runs for human sequences and other biological groups took 36 hours each. Helix-mRNA combines Mamba-2, MLP and attention layers, uses codon-aware tokenization and handles sequences up to 16,384 tokens. The customer story reports a 146% improvement in predictive accuracy for paired mRNA-ribosome measurements and a 52.5% improvement in mRNA half-life correlation compared with previous versions.

Helical shared Nebius’s account of how their work together helped scale models and accelerate customer experiments. Nebius also reports that Helical delivers in-silico tools in weeks rather than months. The customer story describes plans to expand post-training capabilities and develop a larger next-generation Helix-mRNA model.