Agentic-AI-ready genome-wide poxvirus-host interaction screen refined by a protein language model

It proposes ICARus, a positive-unlabelled read-out refinement framework for hit prioritisation in functional screens. Relative to relying on RNA interference screen read-outs alone, the framework introduces positive-unlabelled learning to refine the read-out. The text presents the framework design and goal at the abstract level, without specific performance numbers.

Integrating protein-protein interaction information derived from a protein language model boosts the discovery of vaccinia virus-host interactions. Compared with analyses based only on screen signal, it adds interaction information from a protein language model as an evidence source. The text states the boost as 'boosts the discovery', without giving a specific effect size or control details.

The approach enhances the identification of human genes with potential antiviral function. Beyond screen hits, it points toward human genes that may have antiviral function, serving the discovery of therapeutically relevant targets. The text states this enhancement at the abstract level, without listing specific genes or validation experiments.

It provides the raw and refined read-outs of a genome-wide screen for vaccinia virus host factors as an agentic-AI-enabled community resource. Relative to publishing conclusions only, it offers both raw and refined read-out layers for community reuse and reanalysis. The text explicitly states that 'raw and refined read-outs' are provided, but does not describe data scale or access details.

The work targets research settings that screen host genes involved in infection at the single-cell level using RNA interference and seek to reduce the impact of off-target effects and assay noise; it applies to the discovery of vaccinia virus-host interactions and human genes with potential antiviral function. The proposed ICARus framework is described as generalisable to robust hit prioritisation in functional screens across complex biological systems. The released raw and refined read-outs are intended for the community, including agentic-AI-enabled workflows, to reuse the genome-wide screen data.

The loaded text is at the abstract level and contains no figures, specific effect sizes, control settings, validation experiments, or data scale, so the magnitude of the boost, the content of the hit list, and the concrete form of the refined read-outs cannot be assessed; these remain open questions for readers who go deeper.