Accurate inference of tree topologies from multiple sequence alignments using deep learning

Each directory contains Training, Validation and Test datasets in fasta format. Directory names contain the following information about the MSAs:
1) Gap (w/ indels) or nogap (w/o indels) MSAs2) Number of MSAs per topology generated for training (i.e. 50k MSA * 3 possible topologies = 150k MSAs in TRAIN.fasta file) Order of topologies: 50k x ((D,A),B,C); 50k x ((D,B),A,C); 50k x ((D,C),A,B);3) MSA length (500, 1000 or 10000)4) archive3 contains Keras models and its associated trained weights. They will generate identical CNN accuracies reported in Fig. 2
List of all directories:archive1_gap.tar.gz gap50k_1000 gap50k_10000 gap50k_500archive2_nogap.tar.gz nogap150k_1000 nogap300k_1000 nogap50k_1000 nogap50k_10000 nogap50k_500archive3_trained_keras.tar.gz
gap50k_1000
nogap150k_1000nogap300k_1000nogap50k_1000