← Back to Home

Tracking RNA Therapy Effects Across More Than 10,000 Samples: PacBio and TychoBio Partner to Build AI Training Data

Once an antisense nucleic acid enters a cell, which RNAs does it change? The two companies plan to use full-length sequencing to connect drug design with cellular responses and build predictive tools for rare disease therapies; whether the models can accurately assess efficacy and unintended effects remains to be validated.

By SURL BioNews

Designing RNA therapies for rare diseases involves more than finding a sequence that can act on its target. Researchers also need to understand how a drug candidate changes overall RNA activity after entering a cell. PacBio and TychoBio announced a collaboration on October 6, with plans to use large-scale experiments and full-length RNA sequencing to build datasets linking drug design with cellular responses for artificial intelligence models to learn from.

According to the collaboration announcement, TychoBio plans to use PacBio’s HiFi and Kinnex long-read RNA sequencing technologies to generate data from more than 10,000 samples. Reports on the same announcement by RTTNews and BioTuesdays also confirmed this scale and the planned technologies. This is the number of experimental samples scheduled for analysis; it should not be interpreted as 10,000 patients, nor does it mean the dataset has already been completed.

The first phase will focus on steric-blocking antisense oligonucleotides (SBOs), pairing the designs of different candidate molecules with the transcriptomic changes they cause across multiple cell types. The transcriptome is the overall landscape of the various RNAs within a cell; researchers hope to use it to observe changes brought about by treatment and identify previously unanticipated effects. The two companies also plan to extend this approach to small interfering RNA (siRNA) and other compounds that target RNA.

The value of full-length data lies in preserving clues about different RNA versions and splicing patterns. BioTuesdays cited an explanation by TychoBio founder and CEO Felix Raimundo that long-read sequencing can capture complete transcripts, helping researchers observe RNA isoforms and splicing changes associated with rare diseases, as well as how candidate therapies might alter these processes. For this collaboration, RNA structure and composition after drug action will be important elements for the models to learn from.

The two companies hope the trained models will be able to predict the potency of candidate therapies, their duration of action, and unintended transcriptomic effects across cell types. If these predictions are subsequently validated, they could help researchers decide which designs merit another round of experiments. However, the current announcement describes plans for data generation and model development and does not yet provide model accuracy, independent validation results, or evidence of efficacy in human clinical trials.

Data volume alone also cannot establish whether a model is applicable to a particular disease. The announcement does not list the rare diseases to be prioritized, details of the cell types, or the validation design. Whether responses learned by the models from experimental samples can translate to patients remains a question that subsequent development must answer. In addition, unintended RNA changes can provide research leads, but cannot be directly equated with demonstrated toxicity or clinical safety.

PacBio identifies data generation and the benefits for therapy development as forward-looking plans and states that its products are for research use only, not for use in diagnostic procedures. The concrete step taken by this collaboration is to establish a workflow for RNA drug design that can accumulate experimental evidence. Becoming a reliable tool for predicting treatment effects will still require model validation and subsequent drug research to follow step by step.

References

  1. PacBio
  2. PacBio via GlobeNewswire
  3. RTTNews
  4. BioTuesdays