Source-linked AI summary
AMP Atlas: A Browser-Only Workbench for Antimicrobial Peptide Exploration and Interactive Helical-Wheel Design
Truong-Son Hy
TL;DR
AMP research combines sequence records, chemical modifications, physicochemical descriptors, literature annotations, and experimental structures, and moving among them can require several tools and repeated sequence transfers. AMP Atlas integrates a portable AMP data snapshot, transparent local calculations, molecular inspection, sequence-guided structural comparison, and incremental constraint-based helical-wheel search in a static browser application. At fixed length 23 with no property filters, constraining more positions reduced matching source records from 65 to 1, reaching 1 record at 12 constrained positions.
Problem
AMP research combines sequence records, chemical modifications, physicochemical descriptors, literature annotations, and experimental structures, and moving among them can require several tools and repeated sequence transfers.
Source: §1 · p. 1Method
Results
At fixed length 23 with no property filters, constraining more positions reduced matching source records from 65 to 1, reaching 1 record at 12 constrained positions.
Source: §6.1 · p. 7Takeaways & Limitations
The available data, transparent algorithms, and numerical verification support reproducible exploration with a portable research tool.
Source: §9 · p. 8The evaluation establishes software consistency and reproducibility for a fixed snapshot, but includes no prospective synthesis, biological assay, retrieval-relevance judgment, or comparison of discovery outcomes against another platform.
Source: §7 · p. 8
Abstract
from arXiv · showhide
Accessible research software can connect published antimicrobial peptide (AMP) records to sequence analysis, molecular visualization, and interactive design without requiring a dedicated application server. We present AMP Atlas, an open-source, browser-only workbench that combines a DRAMP-derived database with local physicochemical calculations, published-neighbor retrieval, and sequence-guided structural comparison. The reported snapshot contains 12,784 source records, 10,579 canonical records, 5,183 source SMILES annotations, and 433 PDB chain structures. An inverse helical-wheel editor lets users specify peptide length, fill residues gradually, and immediately retrieve exact-length database matches under positional and property constraints. Selected-position residue frequencies support further editing while preserving the remaining constraints. A Web Worker performs searches and numerical calculations; a WebGL viewer provides interactive molecular representations. The static distribution occupies 138.8 MB and supports GitHub Pages hosting. Verification includes 23 Python tests, 30 browser-core/worker checks, a designer DOM integration suite, and agreement against fixed numerical references. This report describes the implemented algorithms and assumptions; it does not establish predictive accuracy for antimicrobial activity or experimentally validate designed peptides. The website is available at https://hysonlab.github.io/amp-atlas/ and the source code and database are available at https://github.com/HySonLab/amp-atlas
1 Introduction
AMP Atlas addresses the work of moving among AMP sequence, chemical, literature, and structure representations by integrating exploration and design in a browser workbench. Its contribution is an incremental helical-wheel search for published records, not a new biochemical descriptor or activity predictor.
1 Introduction
- Researchers moving among AMP sequence records, chemical modifications, physicochemical descriptors, literature annotations, and experimental structures may need several tools and repeated sequence transfers.The workbench aims to make published results easier to explore, compare, and communicate.Source: §1 · p. 1
- AMP Atlas uses a downloaded DRAMP general-record collection and source SMILES annotations, but neither reproduces every DRAMP collection nor replaces its curation.A canonical sequence alone may omit chemical features affecting peptide properties.Source: §1 · p. 1
- AMP Atlas adds an inverse helical-wheel interaction: users fill numbered positions, inspect the linear sequence, and retrieve published records compatible with partial constraints.The context contrasts this with classical helical-wheel projections, hydrophobic-moment quantification, and HELIQUEST analysis.Source: §1 · p. 1
- The browser workbench integrates a portable data snapshot, transparent calculations, molecular views, and incremental constraint-based search for exploratory research and hypothesis preparation.It is an implementation report, not a new descriptor, alignment algorithm, or learned predictor; it contains no trained activity, toxicity, MIC, helicity, or structure-prediction model.Source: §1 · p. 1
2 System overview and scope
AMP Atlas is a static public website in which searches and scientific calculations run on the visitor’s device. Its connected workflows link record exploration, sequence analysis, structure comparison, wheel editing, and data downloads while processing user inputs locally.
2 System overview and scope
- The public edition serves HTML, CSS, JavaScript, JSON, PDB coordinates, and downloadable files through GitHub Pages, with calculations and searches on the visitor’s device.The hosted edition requires no application server, account, remote inference endpoint, or visitor API key; a local Python implementation is optional.Source: §2 · p. 1
- Five connected workflows cover source-record browsing, sequence analysis, structure comparison, helical-wheel editing, and data downloads, with completed designs entering sequence analysis and published results returning to the explorer.This reuse keeps literature evidence and design inputs connected.Source: §2 · pp. 1–2
- Local scientific processing avoids transmitting input sequences and uploaded PDB contents through the application, but does not constitute a formal privacy audit or guarantee fully offline use.Hosting providers may log ordinary page and public-file requests, external publication links contact their providers, and workers and file fetches require HTTP(S).Source: §2 · p. 2
3 Data collection and representation
AMP Atlas assembles a bounded DRAMP-derived record and structure snapshot, retaining source information while applying explicit rules for canonical sequences, annotations, structure links, and static distribution.
3.1 Data sources and collection
- The snapshot combines official DRAMP general-record and general-SMILES spreadsheets downloaded on 17 October 2026 with referenced PDB files obtained through RCSB PDB.Imported-record and stored-chain counts are not counts of unique chemical entities; 11,718 distinct source sequence strings include noncanonical strings, while canonical records contain 10,056 distinct sequences.Source: §3.1 · p. 2
3.2 Validation and chemical annotations
- Source sequences are retained, while a separate canonical flag requires 1–500 residues over the 20-letter alphabet; noncanonical records remain browseable and downloadable but are not converted for calculations.Recognized terminal amidation annotations inform precomputed descriptors; unknown status is approximated with free termini, and source chemistry is not reconstructed or validated from SMILES.Source: §3.2 · p. 2
3.3 Structure acquisition and linking
- The referenced PDB collection yields 433 stored chains and 315 source records flagged with a local structure, using exact-sequence or qualifying contiguous-fragment matching.The 591 attempted referenced downloads included two failures; links do not establish terminal chemistry, modifications, stereochemistry, or assay conditions, and missing coverage may reflect collection or parser rules.Source: §3.3 · p. 2
3.4 Storage and distribution
- 138.8 MB is the approximate complete static distribution size, while the 9.33 MB search index and 0.95 MB structure index load separately from detailed records and atom files.Annotation shards load on demand, and transfer compression depends on the host.Source: §3.4 · p. 2
4 Interface and research workflows
AMP Atlas connects record exploration and molecular viewing with sequence-based analysis and an inverse helical-wheel designer. The designer treats partial sequences as search constraints and exposes matching records and database frequencies as users edit.
4.1 Record exploration and molecular visualization
- The explorer searches record identifiers, names, sequences, and organism text, while selected records expose descriptors, activity annotations, references, modifications, and available SMILES.The 3Dmol.js viewer renders stored PDB atoms with interactive representations and labels associated structures as sequence-matched references, not structures of new designs.Source: §4.1 · p. 3
4.2 Sequence workbench
- Users can request sequence-based or descriptor-based neighbors after choosing pH and terminal amidation, receiving descriptor cards, composition, hydropathy, charge–pH curves, ideal-helical projections, alignments, and source links.Up to 15 distinct-sequence neighbors are returned according to the selected numerical rule, not as predicted successful analogues.Source: §4.2 · p. 3
4.3 Inverse helical-wheel designer
- The inverse wheel accepts 1–72 residues, preserves numbered N-to-C sequence order, and searches after a 250 ms editing pause for exact-length matches under specified residues and property filters.Selected-position counts reflect records compatible with other constraints, but duplicates, source selection, and publication bias affect these database frequencies; they are not activity probabilities or recommendations. Partial-entry summaries count assigned basic, acidic, and hydrophobic residues without extrapolating full-sequence properties.Source: §4.3 · p. 3
- Figure 2 presents vector reconstructions using magainin 2 and stored PDB 2MAG chain A [8], not browser screenshots; its molecular panel is an illustrative Cα projection, whereas the live site uses an all-atom viewer [7].The illustrated 23-residue sequence does not imply that every deposited residue or chemical group is represented in the descriptor model.Source: Figure 2 · p. 4
5 Algorithms
AMP Atlas computes sequence-derived descriptors under explicit chemical and numerical assumptions, retrieves eligible published neighbors, and compares structures through sequence-guided rigid fitting. Its inverse wheel search matches partially specified sequences against exact-length records while preserving active constraints.
5.1 Sequence-derived descriptors
- Descriptor calculations report mass, charge, pI, hydropathy, hydrophobic moment, composition, hydrophobic fraction, aromaticity, and reduced-cysteine extinction using specified sequence-based models.The canonical linear-chain mass model omits disulfide oxidation, cyclization, nonstandard residues, N-terminal modifications, and adducts; charge uses independent Henderson–Hasselbalch terms and fixed pK values.Source: §5.1 · p. 4
- The hydrophobic moment uses unnormalized Kyte–Doolittle values, so its scale depends on the hydropathy table and neither it nor a wheel predicts helicity.Hydrophobic fraction counts A, V, I, L, M, F, W, and Y; aromaticity uses F/W/Y, and the reduced-cysteine extinction estimate excludes cystine terms.Source: §5.1 · p. 4
5.2 Published-neighbor eligibility
- Similarity and wheel searches require canonical records with source-study annotations, while exact-sequence deduplication returns one ranked representative without merging differing chemistry or evidence.Eligibility uses a PubMed token or sufficiently long title and reference, excluding placeholder titles; the reported heuristic admits 10,489 canonical records.Source: §5.2 · pp. 4–5
5.3 Global sequence similarity
- Sequence retrieval ranks candidates by unit-cost Levenshtein distance, a global minimum-edit alignment computed in O(|s||t|) time per candidate with O(|t|) working memory.Full alignments retain a traceback table and prefer diagonal steps on ties; identity includes gap columns, whereas sequence coverage divides aligned residue-pair count by query length.Source: §5.3 · p. 5
5.4 Physicochemical neighbor score
- The five-vector descriptor score uses fixed engineering scales rather than learned weights or dataset standard deviations, and similarity is not calibrated to functional equivalence.Retrieval query descriptors always use pH 7; candidate descriptors and wheel filters use stored snapshot values, including existing numerical rounding.Source: §5.4 · pp. 5–6
5.5 Wheel geometry and inverse mapping
- The wheel preserves sequence order and deterministically linearizes filled indexed slots; additional rings prevent visual collisions and are not physical helix dimensions or cyclization constraints.Source: §5.5 · p. 6
5.6 Incremental constraint-based retrieval
- Partial-wheel retrieval matches eligible records of the specified length at every filled position, applies active property bounds and optional structure filters, and reports counts before and after filtering.Selected-position frequencies are computed after wildcarding that slot and scanning records; a generation counter ignores stale responses, while debouncing does not stop already-running worker computations.Source: §5.6 · p. 6
5.7 Sequence-guided structure comparison
- Structure comparison aligns target and query sequences to define residue correspondences, then uses a proper rigid fit without reflection or scaling and ranks results by RMSD.The browser uses Horn’s quaternion matrix; the comparison is sequence-guided, not sequence-independent fold search, flexible superposition, or evidence of shared biological function.Source: §5.7 · p. 6
6 Reproducible examples and verification
The examples show how progressively specified sequences narrow matches in the bundled snapshot, alongside calculated descriptors and a sequence-guided structural comparison. Verification checks software behavior and numerical agreement, while execution timings are illustrative rather than browser-performance evidence.
6.1 Progressive search example
- 65 source records match a one-residue prefix, falling to 1 at 12 specified residues; after GIGK, position 5 has counts F:6, H:2, W:1, and Y:1.These are exact worker outputs for the bundled snapshot; a separate pattern with G at position 1 and K at position 4 yields two records under linked-structure and nonnegative-charge filters.Source: §6.1 · p. 7
- Under free-terminus, pH-7 assumptions, the example’s calculated mass is 2466.93 Da, charge is 3.000, GRAVY is 0.083, hydrophobic fraction is 43.5%, pI estimate is 10.501, and ideal-helical Kyte–Doolittle moment is 1.456.For stored query 2MAG_41, the closest returned structural target is 2FLY_41, with RMSD 1.1335 Å and 87.0% coverage; the reference uses the first-model observed canonical trace from PDB 2MAG, whose NMR study used membrane-mimicking conditions.Source: §6.1 · p. 7
6.2 Software and numerical verification
- The verification suite includes 23 Python tests, 30 browser-core/worker checks, and a designer DOM integration suite, but does not validate real-browser rendering or WebGL behavior.Static integrity checks cover all 145 record shards, structure paths, database-copy equality, and asset paths; no visual cross-browser or WebGL compatibility study is claimed.Source: §6.2 · p. 7
- Across 20 fixed fitting fixtures, rounded RMSD outputs agree exactly between implementations, with a maximum absolute fitted-coordinate discrepancy of 3.72 × 10−14 Å.The tests also reject a reflected configuration as a zero-error proper rotation; these checks support numerical consistency between implementations, not external biochemical accuracy.Source: §6.2 · p. 7
6.3 Illustrative execution times
- Table 2 reports warm worker execution times in milliseconds from Node.js measurements, not browser interaction or network latencies.The measurements use one machine and a small number of repetitions, exclude loading, rendering, network transfers, and the 250 ms debounce, and do not support hardware-independent responsiveness claims; exhaustive global sequence comparison is substantially slower than wheel filtering at this scale.Source: Table 2 · p. 7, §6.3 · p. 7
7 Limitations and future work
The workbench’s scope is bounded by its canonical-sequence calculations, selected structure corpus, and fixed-snapshot evaluation. Larger-scale deployment and learned models remain future work with additional requirements.
Chemical and biological scope
- Canonical linear-chain descriptors do not fully represent modified, cyclic, branched, or stereochemically altered peptides, and source SMILES is preserved but not interpreted by the calculator.Assay annotations remain source text without normalization of MIC units, organisms, media, toxicity protocols, or uncertainty; matches do not validate designs, and missing matches do not establish novelty or inactivity.Source: §7 · p. 7
Structural interpretation
- The structure corpus covers a small fraction of source records, selected through available DRAMP references, download limits, and sequence-linking criteria.First-model observed traces omit ensemble variability and unobserved residues; rigid sequence-guided RMSD depends on correspondence, length, and coverage, and unrelated short helices can superpose well.Source: §7 · pp. 7–8
Evidence and evaluation
- Evaluation establishes software consistency and reproducibility for a fixed snapshot, not prospective synthesis, biological assays, user studies, retrieval relevance, or comparative discovery outcomes.Duplicate source records influence selected-position frequencies, which reflect the particular collection and its redundancy.Source: §7 · p. 8, Figure 4 · p. 8
Scale and deployment
- The 138.8 MB static deployment does not establish scalability to much larger data: each visitor’s worker loads the search index, and sequence retrieval scans eligible records.Full alignments require quadratic tables for selected candidates; future larger deployments should measure browser memory, interaction latency, cache behavior, and device variability.Source: §7 · p. 8
AI extensions
- Future learned models would require licensed training data, normalized labels and chemistry, sequence-aware train/test separation, calibrated uncertainty, and independent evaluation.The report proposes keeping such models separate from deterministic descriptors and source frequencies; these are development directions, not implemented results.Source: §7 · p. 8
8 Availability and reproducibility
AMP Atlas is available as a public website and local static-server application, with source code, data, and reproducibility materials distributed for reuse.
Availability
- The public website, source code, and data are available through the project’s public links, and the application can also run locally using a static HTTP server.SQLite, JSON, CSV, FASTA, and FASTS downloads support downstream analysis; FASTS exports retain source SMILES and source FASTS text when present, without independently validating encoded chemistry.Source: §8 · p. 8
Reproducibility
- Supplementary materials provide data provenance, exact query outputs, numerical evaluation results, and scripts for reproducing analyses and figures.The repository includes build tools and verification suites, and the evaluation harness runs the website’s scientific worker against locally stored data.Source: §8 · p. 8
Licensing
- The application source is MIT licensed, while DRAMP-derived material retains upstream CC BY 4.0 terms and PDB archive data follow CC0.AMP Atlas is an independent workbench, not an official DRAMP release.Source: §8 · p. 8
9 Conclusion
AMP Atlas integrates source peptide records, transparent calculations, molecular inspection, and inverse helical-wheel editing in a browser-only workbench. Its available data and numerical verification support reproducible exploration, while experimental assessment, chemical generalization, and predictive modeling remain future work.
Conclusion
- The live designer treats partial sequences as explicit constraints and exposes source records and selected-position counts behind its feedback.The paper’s current contribution is a portable exploratory research tool, not experimental validation or predictive modeling.Source: §9 · p. 8
A Calculation parameters
Table 3 provides the canonical residue parameters used by the implementation, including average residue masses and Kyte–Doolittle hydropathy values. Charge calculations use fixed pK values without experimental fitting.
A Calculation parameters
- Table 3 lists the complete canonical residue parameter tables used by the implementation.Mass values are average residue masses, not free amino-acid or monoisotopic masses.Source: A Calculation parameters · p. 9, Table 3 · p. 9
- The hydropathy column reproduces the Kyte–Doolittle scale.Table 3 identifies h as the Kyte–Doolittle hydropathy value.Source: A Calculation parameters · p. 9, Table 3 · p. 9
- Charge calculations use fixed pK values, and no experimental fit was performed for this report.The listed values include pK values for the termini and residues K, R, H, D, E, C, and Y.Source: A Calculation parameters · p. 9
B Figure provenance
The report’s figures use the current interface geometry and actual worker output. Supplementary sources include exported designer-wheel SVGs and a molecular trace derived from local PDB coordinates.
B Figure provenance
- The report’s figures are generated from the current interface geometry and actual worker output.No source-paper figure is reproduced.Source: B Figure provenance · p. 9
- Original wheel SVGs exported from the designer DOM are included as supplementary figure sources.Source: B Figure provenance · p. 9
- The example molecular trace is derived from local PDB 2MAG chain A coordinates, with the original study cited in the caption.Source: B Figure provenance · p. 9