Fusion2AI Annotations

formatted_cafa_haiku35text_opus5go_predictions_train contains Fusion2AI predictions for the bacterial sequences in the training set for the CAFA 6 Protein Function Prediction Challenge (https://www.kaggle.com/competitions/cafa-6-protein-function-prediction) fusion_go_and_text_sample includes GO terms, raw name lists, and LLM-summarized function descriptions for functions included in our evaluations. The columns are as follows:1. "function_id": the Fusion function ID (e.g., "F-1")2. "cafa_sample_id": the protein ID of the sampled CAFA protein, if applicable3. "swiss_sample_id": the ID of the sampled protein from UniProtKB/Swiss-Prot, if applicable4. "method": the method used for text-generation or GO extraction (possible values are sub-bullets) "CAFA-GT": ground truth from CAFA 6 challenge "Swiss-GT": ground truth GO annotations for the "swiss_sample_id" protein "name_list": the pipe-separated (" | ") name list fed as input to the LLM for function summarization "Haiku-3.5-Text": function summary generated by Haiku 3.5 "Sonnet-4.5-Text": function summary generated by Sonnet 4.5 "Opus-4.6-Text": function summary generated by Opus 4.6 "Haiku-3.5-Text-Opus-5-GO": GO terms extracted with Opus 5 from Haiku-3.5-generated function summaries; includes GO terms across all LLM-estimated reliability levels (low, medium, and high) "Haiku-3.5-Text-Opus-5-GO-high": GO terms extracted with Opus 5 from Haiku-3.5-generated function summaries; includes only GO terms with high reliability, as estimated by Opus 5 during extraction "Haiku-3.5-Text-Opus-5-GO-medium-high": GO terms extracted with Opus 5 from Haiku-3.5-generated function summaries; includes only GO terms with estimated (by Opus 5 during extraction) reliability of medium or high "Haiku-3.5-Text-Opus-4.6-GO": GO terms extracted with Opus 4.6 from Haiku-3.5-generated function summaries "Haiku-3.5-Text-Sonnet-4.5-GO": GO terms extracted with Sonnet 4.5 from Haiku-3.5-generated function summaries "Haiku-3.5-Text-Haiku-4.5-GO": GO terms extracted with Haiku 4.5 from Haiku-3.5-generated function summaries "Opus-4.6-Only": GO terms extracted with Opus 4.6 from Opus-4.6-generated function summaries "Sonnet-4.5-Only": GO terms extracted with Sonnet 4.5 from Sonnet-4.5-generated function summaries "Swiss-Fusion": GO terms from Swiss-Prot sequences aligned with highest HFSP>=14 to Fusion reference sequences "Swiss-Fusion-Experimentally-Validated": experimentally validated GO terms from Swiss-Prot sequences aligned with highest HFSP>=14 to Fusion reference sequences "Swiss-All-Matches": GO terms from all aligned Swiss-Prot sequences to the corresponding CAFA sequence with HFSP>=14 "Swiss-Best-Match": GO terms from the Swiss-Prot sequence aligned to the corresponding CAFA sequence with highest HFSP>=145. "go_namespace": if the row includes GO terms, this is the namespace of those GO terms ("molecular_function", "biological_process", or "cellular_component")6. "go_terms": list of GO terms generated or extracted via the corresponding method separated by "; "7. "text_type": whether the text is an LLM-generated function summary ("llm_function_summary") or a name list fed as input to the LLM ("name_list")8. "text": text for the corresponding method; empty for rows with GO terms fusion_swiss_prot_alignment contains the Swiss-Prot IDs aligned with highest HFSP>=14 to NR100 reference sequences for Fusion functions. Columns are as follows: "fusion_function_id": the Fusion function ID (e.g., "F-1"), "swiss_prot_id": the identifier of an SP protein that is a match with highest HFSP>=14 for at least one sequence in the function's NR100 reference set