PacBio and TychoBio plan AI training data from more than 10,000 RNA samples

PacBio and TychoBio plan to use long-read RNA sequencing to build datasets from more than 10,000 samples, supporting AI research into RNA therapies for rare diseases. The collaboration aims to connect the design of candidate treatments with their effects across the transcriptome, including potential unintended effects.

Watch Desk analysis

What happened

The companies plan to use PacBio HiFi and Kinnex sequencing to measure how steric blocking antisense oligonucleotides affect RNA across multiple cell types. Drug Target Review reports that the resulting data will be used to train AI models to predict which compounds might work, how long their effects could last and where unintended effects may occur.

The partners say they may extend the approach to other RNA-targeting compounds, including small interfering RNA. The announcement describes a research and data-generation effort, not a tested treatment or evidence that the models can yet make reliable predictions. Read Drug Target Review’s report.

Why it matters

RNA therapies can have effects that vary across cell types and transcript forms. The planned dataset is intended to capture those changes, giving researchers material to train models that could help assess candidate therapies before they progress towards clinical trials.

That makes the partnership a concrete example of AI drug discovery starting with experimental biology data, rather than a model being asked to conjure its way past a shortage of evidence. Whether the data ultimately improves therapy selection remains to be shown.

Our read

The useful part here is the proposed link between treatment design and measured, full-transcriptome effects. If the dataset captures enough variation across cell types, it could help researchers examine both potential efficacy and unintended effects earlier in development. For now, this is a research plan with an ambitious dataset target, not a therapeutic breakthrough. The interesting test will be what the resulting models can predict, and whether later experiments bear those predictions out.

What to watch

• Whether the partners reach their target of more than 10,000 samples.
• What data and model results they publish, including validation against experiments.
• Whether the approach expands to siRNA or other RNA-targeting compounds.
• Whether any candidate therapies advance towards clinical testing.

Discussion spark: Should AI drug-discovery projects be judged first by the scale and quality of their training data, or by how soon their predictions hold up in experiments?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.