Oral Peptide Design Landscape
1,879 peptideschemistrydrug-discoverypeptides
A TMAP of an evolutionary peptide-design landscape: cyclic analogues of five oral peptide drugs plus the reference drugs themselves. Each molecule is encoded as MAP4 atom-pair shingles, MinHashed into signatures, and fed through an LSH forest directly into TMAP as a precomputed kNN graph - bypassing TMAP's internal MinHash step entirely, since the data is already hashed. This is the power-user path: build your own index, then hand TMAP the graph. Color layers show target family, fitness score, generation number, and which points are the original reference drugs versus evolved analogues.
Oral Peptide Design LandscapeOpen full page
Loading interactive demo…
How it was made
generate.pypython
from tmap import TMAP
from tmap.index import LSHForest
from tmap.index.encoders.minhash import MinHash
# Shingles already extracted (MAP4 atom-pairs); MinHash them ourselves
mh = MinHash(num_perm=512, seed=42)
signatures = mh.batch_from_string_array(shingle_lists)
forest = LSHForest(d=512)
forest.batch_add(signatures)
forest.index()
knn = forest.get_knn_graph(k=20, kc=50)
# Feed the precomputed graph straight into TMAP - skips internal MinHash
model = TMAP(metric="jaccard", n_neighbors=20, seed=42).fit(knn_graph=knn)
viz = model.to_tmapviz()
viz.add_color_layout("fitness", fitness_scores, categorical=False, color="RdYlBu_r")
viz.add_color_layout("generation", generation, categorical=False, color="viridis")
viz.write_html("oral_peptides.html")