the lab

Phase 1 of 3

Read this first. Everything on this page is computational. No candidate here has been validated, synthesised, or tested in any living system. Nothing here is a drug, a product, or a claim that any sequence does anything. No patent has been filed on any of it, and none will be — the point of the registry is the opposite.

Where this actually is

Phase 1 — running now. Read the indexed corpus, model residue statistics per class, sample and filter novel sequences, hash and publish them. That is what the code below does, and it is all it does.
Phase 2 — not built. Mining single-cell expression atlases and receptor-ligand databases directly for target signatures, rather than working from the peptide corpus alone.
Phase 3 — not built. Structural scoring of candidates against named receptor targets.

Saying so plainly costs nothing. A dashboard pretending to be phase 3 would be found out in an afternoon.

The method, in full

The run below is reproducible: bun run data/generate-candidates.ts with a fixed seed produces exactly these sequences, every time.

Read the 100-entry corpus; keep the 62 entries with an established sequence.
For a target class, build an order-2 residue Markov model from that class's sequences, backing off to the whole-corpus model for unseen contexts.
Sample candidates inside the class's observed length range.
Reject any candidate that is a substring of a known peptide, contains one, or sits within 0.5 normalised edit distance of any indexed sequence.
Reject any candidate whose net charge or mean hydropathy falls outside the class's observed ranges.
Score what remains on composition similarity to the class centroid and on novelty; hash the survivors.

What the score is not. The composite is a weighted blend of composition similarity, novelty and physicochemical range fit. It is not a prediction of binding, activity, safety or efficacy. No such prediction is made anywhere on this site.

Run log

run started 2026-09-01T15:03:11.236Z
read 100 indexed peptides and 761 citations
classes targeted: healing immune nootropic longevity mitochondrial
candidates surviving all filters: 11

Candidates (11)

COMPUTATIONAL CANDIDATE — NOT VALIDATED, NOT SYNTHESISED, NOT FOR HUMAN USE

BLLM-C001

LKKTETIEQEKQAGLVQPGKPDMAEIEKQAGEPPG

healing 35 residues 3776 Da computational score 0.824

designed against: tissue-repair peptides in the corpus (actin-binding and matrix-signalling fragments) — 5 indexed sequences.

reasoning trace

Built an order-2 residue Markov model from the 5 sequenced healing peptides in the corpus.
Sampled a 35-residue candidate inside that class's observed length range (6–40).
Residue composition matches the class centroid at cosine 0.863.
Nearest known sequence is Thymosin β4 at normalised edit distance 0.698 — above the 0.5 novelty floor, and not a substring of any indexed peptide.
Net charge -2 and mean hydropathy -1.12 both fall inside the class's observed ranges.

composition 0.863 · novelty 0.698 (nearest known: Thymosin β4) · charge in class range: yes · hydropathy in class range: yes

sha256 9839d67d39e13c20800667e53518ac5974449bdaf369442971e0b23d37d6faae

registry entry →

COMPUTATIONAL CANDIDATE — NOT VALIDATED, NOT SYNTHESISED, NOT FOR HUMAN USE

BLLM-C002

LKKTETIEQEKQAGLVQPGKPD

healing 22 residues 2438 Da computational score 0.829

designed against: tissue-repair peptides in the corpus (actin-binding and matrix-signalling fragments) — 5 indexed sequences.

reasoning trace

Built an order-2 residue Markov model from the 5 sequenced healing peptides in the corpus.
Sampled a 22-residue candidate inside that class's observed length range (6–40).
Residue composition matches the class centroid at cosine 0.890.
Nearest known sequence is TB-500 at normalised edit distance 0.682 — above the 0.5 novelty floor, and not a substring of any indexed peptide.
Net charge 0 and mean hydropathy -1.25 both fall inside the class's observed ranges.

composition 0.89 · novelty 0.682 (nearest known: TB-500) · charge in class range: yes · hydropathy in class range: yes

sha256 bc96f27e9699f6b17326a73bb23cfd66f472291afac4a02bd14bf9929eccb7e2

registry entry →

COMPUTATIONAL CANDIDATE — NOT VALIDATED, NOT SYNTHESISED, NOT FOR HUMAN USE

BLLM-C003

LKKTETIEKQAGESQGTFTQE

healing 21 residues 2354 Da computational score 0.76

designed against: tissue-repair peptides in the corpus (actin-binding and matrix-signalling fragments) — 5 indexed sequences.

reasoning trace

Built an order-2 residue Markov model from the 5 sequenced healing peptides in the corpus.
Sampled a 21-residue candidate inside that class's observed length range (6–40).
Residue composition matches the class centroid at cosine 0.733.
Nearest known sequence is TB-500 at normalised edit distance 0.667 — above the 0.5 novelty floor, and not a substring of any indexed peptide.
Net charge -1 and mean hydropathy -1.32 both fall inside the class's observed ranges.

composition 0.733 · novelty 0.667 (nearest known: TB-500) · charge in class range: yes · hydropathy in class range: yes

sha256 e302b1994143dfe5d136b71eb70a2cbba081283d64b3ad4bca7e21e2a82b256f

registry entry →

COMPUTATIONAL CANDIDATE — NOT VALIDATED, NOT SYNTHESISED, NOT FOR HUMAN USE

BLLM-C004

LLGDFFRKRIKDFFRKRQQELDKWASLIHSLWNWFTSSEI

immune 40 residues 5044 Da computational score 0.75

designed against: host-defence and immunomodulatory peptides in the corpus — 9 indexed sequences.

reasoning trace

Built an order-2 residue Markov model from the 9 sequenced immune peptides in the corpus.
Sampled a 40-residue candidate inside that class's observed length range (6–40).
Residue composition matches the class centroid at cosine 0.724.
Nearest known sequence is LL-37 at normalised edit distance 0.650 — above the 0.5 novelty floor, and not a substring of any indexed peptide.
Net charge +3 and mean hydropathy -0.59 both fall inside the class's observed ranges.

composition 0.724 · novelty 0.65 (nearest known: LL-37) · charge in class range: yes · hydropathy in class range: yes

sha256 35ca9a3c18716617dcb6c446f2dc6ed84959ecdad4048f685ac0973daf76805b

registry entry →

COMPUTATIONAL CANDIDATE — NOT VALIDATED, NOT SYNTHESISED, NOT FOR HUMAN USE

BLLM-C005

YTSSEITTKDFFRKSGAVDTSLIHSLWNWFAYRKSQNQQE

immune 40 residues 4771 Da computational score 0.804

designed against: host-defence and immunomodulatory peptides in the corpus — 9 indexed sequences.

reasoning trace

Built an order-2 residue Markov model from the 9 sequenced immune peptides in the corpus.
Sampled a 40-residue candidate inside that class's observed length range (6–40).
Residue composition matches the class centroid at cosine 0.761.
Nearest known sequence is CJC-1295 at normalised edit distance 0.750 — above the 0.5 novelty floor, and not a substring of any indexed peptide.
Net charge +1 and mean hydropathy -0.87 both fall inside the class's observed ranges.

composition 0.761 · novelty 0.75 (nearest known: CJC-1295) · charge in class range: yes · hydropathy in class range: yes

sha256 b5ef9c9f2e510b69cef7aa61b2f695005b0753fce65a9a42a2c7e30ad77369f4

registry entry →

COMPUTATIONAL CANDIDATE — NOT VALIDATED, NOT SYNTHESISED, NOT FOR HUMAN USE

BLLM-C006

SDAAVDTSSEITTGLPGTCGLPGTCLKSGAICHPV

immune 35 residues 3360 Da computational score 0.731

designed against: host-defence and immunomodulatory peptides in the corpus — 9 indexed sequences.

reasoning trace

Built an order-2 residue Markov model from the 9 sequenced immune peptides in the corpus.
Sampled a 35-residue candidate inside that class's observed length range (6–40).
Residue composition matches the class centroid at cosine 0.728.
Nearest known sequence is Thymosin α1 at normalised edit distance 0.600 — above the 0.5 novelty floor, and not a substring of any indexed peptide.
Net charge -2 and mean hydropathy 0.30 both fall inside the class's observed ranges.

composition 0.728 · novelty 0.6 (nearest known: Thymosin α1) · charge in class range: yes · hydropathy in class range: yes

sha256 f00089fcf9f84a035b167df750c63b3b280077bcb4f3cb534173c5ea62f29489

registry entry →

COMPUTATIONAL CANDIDATE — NOT VALIDATED, NOT SYNTHESISED, NOT FOR HUMAN USE

BLLM-C007

EDGSFR

longevity 6 residues 710 Da computational score 0.66

designed against: short geroprotective regulatory peptides in the corpus — 3 indexed sequences.

reasoning trace

Built an order-2 residue Markov model from the 3 sequenced longevity peptides in the corpus.
Sampled a 6-residue candidate inside that class's observed length range (6–4).
Residue composition matches the class centroid at cosine 0.650.
Nearest known sequence is Pinealon at normalised edit distance 0.500 — above the 0.5 novelty floor, and not a substring of any indexed peptide.
Net charge -1 and mean hydropathy -1.65 both fall inside the class's observed ranges.

composition 0.65 · novelty 0.5 (nearest known: Pinealon) · charge in class range: yes · hydropathy in class range: yes

sha256 1fee88d0b9ced962d2389e59a4e99736ca21bf43f01b2475b2e14f9393c6f383

registry entry →

COMPUTATIONAL CANDIDATE — NOT VALIDATED, NOT SYNTHESISED, NOT FOR HUMAN USE

BLLM-C008

EDGSFR

longevity 6 residues 710 Da computational score 0.66

designed against: short geroprotective regulatory peptides in the corpus — 3 indexed sequences.

reasoning trace

Built an order-2 residue Markov model from the 3 sequenced longevity peptides in the corpus.
Sampled a 6-residue candidate inside that class's observed length range (6–4).
Residue composition matches the class centroid at cosine 0.650.
Nearest known sequence is Pinealon at normalised edit distance 0.500 — above the 0.5 novelty floor, and not a substring of any indexed peptide.
Net charge -1 and mean hydropathy -1.65 both fall inside the class's observed ranges.

composition 0.65 · novelty 0.5 (nearest known: Pinealon) · charge in class range: yes · hydropathy in class range: yes

sha256 1fee88d0b9ced962d2389e59a4e99736ca21bf43f01b2475b2e14f9393c6f383

registry entry →

COMPUTATIONAL CANDIDATE — NOT VALIDATED, NOT SYNTHESISED, NOT FOR HUMAN USE

BLLM-C009

MAPRKLRIVQQEMGYIFYP

mitochondrial 19 residues 2341 Da computational score 0.789

designed against: mitochondrial-derived peptides in the corpus — 2 indexed sequences.

reasoning trace

Built an order-2 residue Markov model from the 2 sequenced mitochondrial peptides in the corpus.
Sampled a 19-residue candidate inside that class's observed length range (16–24).
Residue composition matches the class centroid at cosine 0.842.
Nearest known sequence is MOTS-c at normalised edit distance 0.632 — above the 0.5 novelty floor, and not a substring of any indexed peptide.
Net charge +2 and mean hydropathy -0.22 both fall inside the class's observed ranges.

composition 0.842 · novelty 0.632 (nearest known: MOTS-c) · charge in class range: yes · hydropathy in class range: yes

sha256 db6ad6c1ec7aa1705577330a114a7591d2842d27c04f7ca8a7c9e364cfac781c

registry entry →

COMPUTATIONAL CANDIDATE — NOT VALIDATED, NOT SYNTHESISED, NOT FOR HUMAN USE

BLLM-C010

MAPRKLRISWIKDFIEWLKNGG

mitochondrial 22 residues 2659 Da computational score 0.796

designed against: mitochondrial-derived peptides in the corpus — 2 indexed sequences.

reasoning trace

Built an order-2 residue Markov model from the 2 sequenced mitochondrial peptides in the corpus.
Sampled a 22-residue candidate inside that class's observed length range (16–24).
Residue composition matches the class centroid at cosine 0.780.
Nearest known sequence is Semaglutide at normalised edit distance 0.710 — above the 0.5 novelty floor, and not a substring of any indexed peptide.
Net charge +3 and mean hydropathy -0.39 both fall inside the class's observed ranges.

composition 0.78 · novelty 0.71 (nearest known: Semaglutide) · charge in class range: yes · hydropathy in class range: yes

sha256 6c1a20d3c5e95e597ebcc8731a4df966a3f642615da88649a69031e0150b4c5a

registry entry →

COMPUTATIONAL CANDIDATE — NOT VALIDATED, NOT SYNTHESISED, NOT FOR HUMAN USE

BLLM-C011

MAPRKLRNLISDAIFYPR

mitochondrial 18 residues 2162 Da computational score 0.797

designed against: mitochondrial-derived peptides in the corpus — 2 indexed sequences.

reasoning trace

Built an order-2 residue Markov model from the 2 sequenced mitochondrial peptides in the corpus.
Sampled a 18-residue candidate inside that class's observed length range (16–24).
Residue composition matches the class centroid at cosine 0.868.
Nearest known sequence is Humanin at normalised edit distance 0.625 — above the 0.5 novelty floor, and not a substring of any indexed peptide.
Net charge +3 and mean hydropathy -0.27 both fall inside the class's observed ranges.

composition 0.868 · novelty 0.625 (nearest known: Humanin) · charge in class range: yes · hydropathy in class range: yes

sha256 e854517a9a39b4f583e32668a9110f04adae8e7287a955ac0504d94084e5877a

registry entry →

The queue

extend class coverage to the melanocortin and incretin classes once more of each carries an indexed sequence
widen the corpus beyond 100 entries — every new indexed peptide sharpens the class models
begin anchoring hashes on Robinhood Chain at launch; the registry column stays empty until a real transaction exists