2026-10-12T07:03:15Z https://api.figshare.com/v2/oai

oai:figshare.com:article/34016912 2026-09-28T17:38:29Z category_13 category_14 category_21 category_734 category_931 category_811 category_133 portal_5 item_type_3 month_year_09_2026

<p>Datasets utilized for evaluation of GAMA. We evaluated GAMA on three types of datasets. (1) 270 synthetic sequences that were systematically created to evaluate GAMA for robustness to noise, motif position invariance, and motif logic. (2) Data from the antibody simulation framework Absolut! [44] to demonstrate the viability of GAMA under complex conditions. For each antigen, we selected the top 1% best binding CDRH3 sequences based on energy and filtered them for sequences with one paratope-binding interaction. (3) Experimental mAb HER-2 binder sequences from the Mason dataset [72].</p> Robert Frank (2035264) Michael Widrich (8210652) Rahmad Akbar (17405644) Günter Klambauer (8287176) Geir Kjetil Sandve (10291229) Philippe A. Robert (7350209) Victor Greiff (561226) Genetics Molecular Biology Biotechnology Biological Sciences not elsewhere classified Information Systems not elsewhere classified Mathematical Sciences not elsewhere classified Plant Biology positively labeled sequences known ground truths antigen binding data biological sequences enriched negative labeled data data </ p negative data biological settings trained solely therapeutic design statistical behavior powerful framework particularly advantageous integrated gradients experimental antibody desirable properties biologically ill attribution methods attribution method attribution assignment affinity antibodies <p>Datasets utilized for evaluation of GAMA. We evaluated GAMA on three types of datasets. (1) 270 synthetic sequences that were systematically created to evaluate GAMA for robustness to noise, motif position invariance, and motif logic. (2) Data from the antibody simulation framework Absolut! [<a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1014805#pcbi.1014805.ref044" target="_blank">44</a>] to demonstrate the viability of GAMA under complex conditions. For each antigen, we selected the top 1% best binding CDRH3 sequences based on energy and filtered them for sequences with one paratope-binding interaction. (3) Experimental mAb HER-2 binder sequences from the Mason dataset [<a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1014805#pcbi.1014805.ref072" target="_blank">72</a>].</p> 2026-09-28T18:03:38Z Dataset Dataset 10.1371/journal.pcbi.1014805.t001 https://figshare.com/articles/dataset/_p_Datasets_utilized_for_evaluation_of_GAMA_We_evaluated_GAMA_on_three_types_of_datasets_1_270_synthetic_sequences_that_were_systematically_created_to_evaluate_GAMA_for_robustness_to_noise_motif_position_invariance_and_motif_logic_2_Data_fr/34016912 CC BY 4.0