@platforma-open/milaboratories.sequence-embeddings 1.1.0 → 1.2.0

package/block-pack/CHANGELOG.md

@@ -0,0 +1,3 @@
+## 1.0.0
+
+Initial release.

package/block-pack/description.md

@@ -0,0 +1,27 @@
+# Overview
+
+Turns antibody, TCR, and peptide sequences into numerical vectors — *embeddings* — that place biologically similar sequences close together, so you can compare, cluster, and visualize a repertoire or library by similarity even when the amino acid sequences differ. The vectors come from a protein language model trained on tens of millions of natural proteins, capturing structural and evolutionary patterns that plain sequence identity misses.
+
+The block reads peptide datasets (from Peptide Profiling) or antibody/TCR clonotypes (from MiXCR Clonotyping or Import V(D)J Data, bulk or single-cell). You choose which region to embed: the **full peptide**; the **CDR3** per chain; the **full variable domain** per chain; the **paired Fv** (heavy and light chains together, antibody only); or the **scFv** construct (the full VH–linker–VL polypeptide). Full-domain and Fv options require a fully assembled variable domain. When the input carries only CDR3, only the CDR3 option is offered.
+
+You also choose the **model** for each embedding: a **universal** protein language model that works on any sequence type, or a **specialist** model tuned to a particular kind of sequence — antibodies, nanobodies, TCRs, or peptides — for sharper representations on that type. The block recommends a suitable model for each sequence you pick and only offers models compatible with it. For the universal model you can additionally choose a **fidelity** — *Standard* (faster) or *High* (higher quality, slower). The block uses a GPU automatically when one is available; the larger models (High-fidelity universal, some specialists) are much faster on a GPU, so on a CPU-only machine with many sequences a lighter model is the best choice.
+
+Embeddings are numeric vectors capturing each sequence's properties — used downstream for clustering and 2D sequence-space maps. In Platforma they feed directly into **Embedding Clustering** and **Sequence Space**, for similarity grouping and visualization.
+
+The **universal** model is **ESM-2**, a protein language model developed by Meta AI Research and released under the MIT license. For more information, please see [https://github.com/facebookresearch/esm](https://github.com/facebookresearch/esm) and cite the following publication if used in your research:
+
+> Lin Z, Akin H, Rao R, Hie B, Zhu Z, Lu W, Smetanin N, Verkuil R, Kabeli O, Shmueli Y, dos Santos Costa A, Fazel-Zarandi M, Sercu T, Candido S, Rives A. Evolutionary-scale prediction of atomic-level protein structure with a language model. _Science_ **379**(6637), 1123–1130 (2023). [https://doi.org/10.1126/science.ade2574](https://doi.org/10.1126/science.ade2574)
+
+The **specialist** models are third-party open-source protein language models, each included under its own permissive, commercially-usable license. If you use a specialist model, please also cite its authors:
+
+> **CurrAb** (antibody) — Burbach SM, Briney B. A curriculum learning approach to training antibody language models. _PLOS Computational Biology_ (2025). [https://doi.org/10.1371/journal.pcbi.1013473](https://doi.org/10.1371/journal.pcbi.1013473)
+
+> **AbLang2** (antibody, paired) — Olsen TH, Moal IH, Deane CM. Addressing the antibody germline bias and its effect on language models for improved antibody design. _Bioinformatics_ **40**(11), btae618 (2024). [https://doi.org/10.1093/bioinformatics/btae618](https://doi.org/10.1093/bioinformatics/btae618)
+
+> **VHHBERT** (nanobody) — Tsuruta H, Yamazaki H, Maeda R, Tamura R, Imura A. A SARS-CoV-2 Interaction Dataset and VHH Sequence Corpus for Antibody Language Models. _NeurIPS_ (2024). [https://arxiv.org/abs/2405.18749](https://arxiv.org/abs/2405.18749)
+
+> **H3BERTa** (antibody CDR-H3) — Rodella C, Lemmin T. H3BERTa: A CDR-H3-specific language model for antibody repertoire analysis. _Patterns_, 101561 (2026). [https://doi.org/10.1016/j.patter.2026.101561](https://doi.org/10.1016/j.patter.2026.101561)
+
+> **TCR-BERT** (TCR) — Wu K, Yost KE, Daniel B, Belk JA, Xia Y, Egawa T, Satpathy AT, Chang HY, Zou J. TCR-BERT: learning the grammar of T-cell receptors for flexible antigen-binding analyses. _bioRxiv_ (2021); PMLR: Machine Learning for Health (2024). [https://doi.org/10.1101/2021.11.18.469186](https://doi.org/10.1101/2021.11.18.469186)
+
+> **PeptideCLM-2** (peptide) — Feller AL, Secor M, Swanson S, Wilke CO, Deibler K. Scaling SMILES-based chemical language models for therapeutic peptide engineering. _bioRxiv_ (2026). [https://doi.org/10.64898/2026.01.06.697994](https://doi.org/10.64898/2026.01.06.697994)

package/block-pack/manifest.json

@@ -0,0 +1 @@
+{"schema":"v2","description":{"id":{"organization":"milaboratories","name":"sequence-embeddings","version":"1.2.0"},"components":{"workflow":{"type":"workflow-v1","main":{"type":"relative","path":"main.plj.gz"}},"model":{"type":"relative","path":"model.json"},"ui":{"type":"relative","path":"ui.tgz"}},"meta":{"title":"Sequence Embeddings","description":"Per-sequence vector embeddings for peptide and antibody/TCR datasets, using universal or sequence-specific specialist language models.","longDescription":{"type":"relative","path":"description.md"},"changelog":{"type":"relative","path":"CHANGELOG.md"},"logo":{"type":"relative","path":"block-logo.png"},"url":"https://github.com/platforma-open/sequence-embeddings","support":"mailto:support@milaboratories.com","tags":["airr","downstream","peptide","embeddings"],"organization":{"name":"MiLaboratories Inc","url":"https://milaboratories.com/","logo":{"type":"relative","path":"organization-logo.png"}},"supportedPlatforms":["linux-x64","macosx-aarch64"]},"featureFlags":{"supportsLazyState":true,"supportsPframeQueryRanking":true,"requiresUIAPIVersion":3,"requiresModelAPIVersion":2,"requiresCreatePTable":2,"requiresPFramesVersion":1001031,"requiresPFrameSpec":true,"requiresPFrame":true,"requiresDialog":true}},"timestamp":1783071699351,"files":[{"name":"main.plj.gz","size":818816,"sha256":"B30ED527D9EE8C11307B1BE3747AFBB7420CDFEFA98C66ABB49A06821549D8ED"},{"name":"model.json","size":335943,"sha256":"0F4B78A0B6DF4392D3588C3081363D70FE7A6CC04468083AE202290950FD808B"},{"name":"ui.tgz","size":3817789,"sha256":"EFC1ED4C7B69E882FCAE8C198774BAE0BEE5196A358DC8D078C63F583C7F3792"},{"name":"organization-logo.png","size":24439,"sha256":"FA71390C77C91E4B7FAAE5640D00F92F1E3F2869296F68B6040DD7CC549A50B5"},{"name":"description.md","size":4586,"sha256":"DEF35BAE94AE0279BD202DADADDA3DA3591BCBD7802733BCFE587F9911374F9D"},{"name":"CHANGELOG.md","size":27,"sha256":"3E1D1DA422E5D609FCDE7F996A81F7635011F3CA297923598DB36D924C4A0DAE"},{"name":"block-logo.png","size":21527,"sha256":"6BB33BAF0CD039549661B51AE490373BE60D1811EC71F5023400928293BC2427"}]}