List of matched MSA records

[psridiyer]

I am trying to figure out the matched records -- I understand it is species-wise MSA matching. (some may include padded MSA)

core.py has the following code (lines 601-602):
list_idx.append(
DCN(torch.einsum("pq,p->q", (gs_matching_mat, target_idx)))
)

During training I see the display (enclosed screen shot). - for two species. One with 2 MSA, another with 25.

I have also enclosed a screen shot of shuffled index (for the last 3 iterations of the training)

Question: Species X has 2 (MSA) seqs. Species Y has 25 seqs. How do I figure which of the 25 seq from chain A (left) got matched to the 25 seq in chain B (right) of Species Y? (I assume the MSA sequences for each species are numbered 0 per the original MSA dataset) Thank you

[psridiyer]

I have a case where 5 different species are utilized. Each matrix is clearly shown. How do I pull the records from the (input_left, input_right) records so the final set of matched pair is available. Thank you.

[psridiyer]

gs_matching_mat / (out put list mats) has the data used to produce the above figures. How to take the index from this matrix to the original input_left / input_right data so I can get the "paired MSA"s. thanks