Screening sequence

Andes hantavirus

Three viral proteins against every drug the FDA has approved, in 33 seconds, and the structures behind the two candidates it puts at the top.

01 · Pathogen

Three viral proteins, scored against every human protein.Andes hantavirus against the human proteome

A six-part sequence built from a shipped computational run. It begins with Synthyra’s predicted human protein-interaction map of 19,982 reviewed proteins, beside which the three reviewed Andes orthohantavirus proteins are placed: the nucleocapsid N protein (UniProt O36307, 428 aa), the L protein RNA-dependent RNA polymerase (Q9E005, 2,153 aa) and the GP glycoprotein precursor (Q9E006, 1,138 aa). Each is drawn with edges to the human proteins Atlas-PPI predicts it engages above score 0.9, which is 143, 104 and 28 partners respectively, and those partner sets carry ontology enrichment: mitochondrial outer membrane protein insertion for N, DNA repair and single-stranded DNA binding for L, and choline transport for GP. The sequence then adds a SwissBioPics human-cell diagram, the FDA-approved drug catalog of 2,638 small molecules and 626 biologics, and the screen itself, drawn as one binned column per compound for each of the three targets. The field is first shown in the order the run scored it and then sorted, which is the ranking. Mupirocin, an FDA-approved topical antibiotic from 1987, lands at rank 7 of 2,638 against the N protein at Atlas-PLI probability 0.980, and dornase alfa, approved in 1993 and delivered by nebulizer to the lung, lands at rank 5 of 626 biologics against the L protein at 0.625. The final beat cycles through both predicted complexes as space-filling molecular surfaces moving along their elastic-network normal modes. Beside each is the whole set of complexes Protenix v2 folded against that target, sorted by interface confidence, with the rank Atlas had given each compound: 14 complexes on the N protein and 15 on the L protein, in each case the ten highest-scoring compounds from the screen plus the four antivirals with documented or in vitro hantavirus activity. The two orderings differ, which is the point of showing both. On the N protein the screen ranked lutein first and a block of six vitamin D analogues behind it, and folding puts mupirocin, its seventh, at the head at ipTM 0.867 and ipSAE 0.144, above remdesivir 0.853 and 0.095, baloxavir marboxil 0.815 and 0.077, favipiravir 0.689 and 0.019 and ribavirin 0.662 and 0.012. On the L protein dornase alfa leads at 0.780 and 0.472 against 0.009, 0.009, 0.000 and 0.000, and that interface is reproduced by ESMFold2, an architecturally distinct folding model, at ipTM 0.732 and ipSAE 0.403.

Info

Where every object on screen came from.

Interaction map: Synthyra Atlas-PPI predicted human interactome, 19,982 reviewed proteins.
Pathogen proteome: Andes orthohantavirus, UniProt proteome UP000204348, taxid 1980456. Three reviewed proteins: O36307 (N, 428 aa), Q9E005 (L, 2,153 aa), Q9E006 (GP, 1,138 aa).
Host-pathogen edges: Atlas-PPI, pathogen-human pairs retained above score 0.9 against the 19,933-protein reviewed human proteome. 143, 104 and 28 partners for N, L and GP.
Enrichment: Enrichr libraries over the retained partner sets, Benjamini-Hochberg corrected within each library. Shown: GO:0045040 for N (adjusted p 4.5e-15, 9 proteins), GO:0006281 for L (1.1e-45, 39), GO:0015871 for GP (9.7e-05, 3).
Drug catalog: Synthyra/FDA-Approved-Drugs. 2,638 approved small molecules screened with Atlas-PLI, 626 approved biologics with Atlas-PPI, withdrawn drugs excluded. The screen took 33 seconds on a single GPU including weight loading and preembedding.
Ranks: Raw sigmoid probabilities over the full pool for each target. Mupirocin is rank 7 of 2,638 on the N protein at 0.980; lutein is rank 1. Dornase alfa is rank 5 of 626 on the L protein at 0.625.
Structures: Predicted complexes of the ANDV N protein with mupirocin and the ANDV L protein with dornase alfa, Protenix v2, MSA-enabled, seed 101, best-ranked sample. The N protein is shown as residues 80 to 428, which holds all 27 of mupirocin’s contact residues. The L protein is shown in full.
Motion: Anisotropic elastic-network normal modes from the C-alpha positions of each complex.
Structural follow-up: Protenix v2 over the top 10 Atlas-PLI and top 10 Atlas-PPI predictions per target plus four antiviral controls against all three targets, for 72 complexes.
The table: Every complex folded against the target on screen, sorted by ipSAE, with the rank Atlas gave each compound: 14 rows for the N protein, 15 for the L protein.
Second folding model: ESMFold2 (esmfold2-2026-05), architecturally distinct from Protenix v2. It reproduces the L protein and dornase alfa interface at ipTM 0.732 and ipSAE 0.403, against Protenix’s 0.780 and 0.472. On mupirocin it gives ipTM 0.259, and still ranks it above all four antiviral controls.
Interface metrics: ipTM is the interface predicted TM-score. ipSAE is the interaction prediction score from aligned errors, reported as ipSAE-max. It scales with interface size, so every row on the table names the pool it was screened in.

Nothing in this sequence is an illustration. Every readout is a prediction, and it shows direction rather than an absolute number. What each object is and where it came from are in the disclosure above.