Datasets:

BenchmarkDatasets
/

avida_hil6_onehot

license: cdla-permissive-2.0
task_categories:
  - tabular-classification
pretty_name: avida_hil6_onehot
tags:
  - tabular
  - proteomics
  - binary
  - Biomedical
  - Benchmark
  - antigenomics
configs:
  - config_name: avida_hil6_onehot
    data_files:
      - split: train
        path: data/train.parquet
      - split: validation
        path: data/val.parquet
      - split: test
        path: data/test.parquet
dataset_info:
  description: >-
    Binary classification: predict VHH nanobody binding to one of 17 hIL6
    antigen variants. Features: padded integer AA token IDs — VHH padded to max
    VHH length, Ag padded to max Ag length, then concatenated; vocab size 21
    (0=PAD). All columns are categorical (cat_idx covers all).. Split:
    mutant-based — test={P42A,T48A,E51A,I57A,I60A,K69A,C78A,S81A},
    val={E87A,L120A,L126A,L129A,Q144A,D162A,T165A}, train=rest (incl. WT).