Fig. 5: Zero-shot mutation prediction. | Nature Communications

Fig. 5: Zero-shot mutation prediction.

From: FusOn-pLM: a fusion oncoprotein-specific language model via adjusted rate masking

Fig. 5

A FusOn-pLM performs zero-shot mutation discovery via its MLM head through sequential unmasking of individual residues. Potential mutations are ranked by their logit values. B FusOn-pLM logits for the longest EWSR1::FLI1, PAX3::FOXO1, and TRIM24::RET sequences in FusOn-DB. Yellow regions are considered highly conserved domains. C Recovery of mutations found to cause drug resistance in patients with EML4::ALK and BCR::ABL1-driven cancers. D Case study on kinase fusion ETV6::NTRK3 (647 amino acids), which drives various cancers. FusOn-pLM predictions of NTRK3 kinase domain mutations identified in ETV6::NTRK3+ cancer patients with drug resistance are shown in the table. Based on logit values, disordered residues from the head protein ETV6 are indicated. Source data for this figure are provided in the Source Data file.

Back to article page