Journey
Therapeutic development
From what the literature already supports, through target and assay work, to the packages a programme needs before it can move. Each stage declares its own inputs, tools, artifacts and review point, and is marked with whether the software exists today.
What this is, and what it is not
Hordago Labs builds software that orchestrates computational workflows. No wet laboratory is operated here. Benchwork, sample handling and instrument runs stay with the laboratory that owns them; this software plans the analysis around that work, runs the computation, and records what was decided and on what evidence.
The stages
The later stages of this journey are the thinnest. Chemistry beyond molecular descriptors, the absorption and toxicity work, the translational stage and the regulatory package have no supporting repository, and are marked planned below rather than described as if they ran.
3 built · 4 partly built · 3 planned · 10 stages in all. Counted from the registry this page renders.
-
01 / 10Built Repositories exist for this stage and carry the code described.
Evidence landscape
Establish what the literature actually supports about a disease area before any target is chosen.
Goes in
- Disease or phenotype in scope
- Publications, preprints and reviews
- Datasets the group already holds
Does the work
- paper-intelligence — paper ingestion, claim extraction, contradiction lookup
- fact-check-os — primary-source refetch, consensus pass, adversarial pass
- biocontext7 — retrieval skill bundles for bioinformatics agents
Comes out
- Claim graph linking each claim to its source locator
- Contradiction report where sources disagree
- Source list a reader can open
Person reviews A scientist reads the claim graph and marks which claims are load-bearing before anything downstream is allowed to rest on them.
How this status was set paper-intelligence, fact-check-os and biocontext7 are all present as repositories under the Hordago Labs organisation, each with source packages, skills and a test directory.
-
02 / 10Built Repositories exist for this stage and carry the code described.
Target and mechanism
Turn a disease area into a short list of targets with a mechanistic story that can be attacked.
Goes in
- Claim graph from the previous stage
- Association summary statistics
- Regulatory and expression data for the locus
Does the work
- GWASos — association analysis and inflation checks
- finemap-os — statistical fine-mapping to credible sets
- varfx-os — variant annotation and effect tiers
- reg-os — enhancer, transcription-factor binding and chromatin accessibility analysis
- research-knowledge-graph — Biolink-schema graph queries and hybrid search
Comes out
- Credible sets and locus plots
- Annotated variant table with effect tiers
- Enhancer-to-gene links
- Ranked target list with the evidence behind each rank
Person reviews The target list is read against the claim graph before it is carried forward; a target whose support is weak is sent back rather than promoted.
How this status was set Every tool in this stage is backed by a repository with Python sources and commit history rather than by a plan. Published product names differ from the names their repositories carry.
-
03 / 10Partly built Part of this stage has a repository behind it; a named part does not.
Assay strategy
Decide what experiment would move the target list, and what result would count as a negative.
Goes in
- Ranked target list
- Model system and readout options open to the laboratory
- Prior screens on related targets
Does the work
- co-scientist — hypothesis generation, critique, debate ranking and experimental planning
- science-runtime — planning packet carrying claim scope and a falsification statement
Comes out
- Ranked, falsifiable hypotheses
- Planning packet stating the claim the experiment would support
- Draft assay outline handed to the assay development journey
Person reviews A principal investigator works through the checkpoint cards, which surface where confidence is low, and accepts or rejects each hypothesis.
How this status was set co-scientist and science-runtime exist and carry the planning and hypothesis work. There is no assay-design repository under the organisation, so the step from an accepted hypothesis to a specified assay is done by the assay development journey rather than by a tool of its own.
-
04 / 10Built Repositories exist for this stage and carry the code described.
Perturbation and screening
Design the perturbation, plan the screen, and analyse the counts that come back from the laboratory.
Goes in
- Target list and cell line
- Delivery context
- Counts matrix returned by the laboratory
Does the work
- CRISPRos — guide design, off-target analysis, delivery and editing-outcome skills
- screen-os — library QC, counting, normalisation, batch and copy-number correction, gene-level scoring
Comes out
- Guide library with an off-target report
- Experiment manifest
- Guide-level and gene-level QC report
- Ranked hit list
Person reviews crispr-os requires an explicit safety-review approval file before its validation and transfection phases. screen-os checks phase criteria automatically, so review there is a person reading the quality-control report rather than the software stopping.
How this status was set crispr-os carries the guide-design, off-target and delivery skills; screen-os carries library quality control, counting, normalisation, batch and copy-number correction and gene-level scoring. Both have Python sources and skill directories for the steps named above.
-
05 / 10Partly built Part of this stage has a repository behind it; a named part does not.
Virtual screening and molecular design
Fold the target, find a pocket, and dock candidate binders against it.
Goes in
- Target sequence or structure
- Candidate ligands as SMILES
- Pocket definition
Does the work
- struct-os — structure prediction, binding-site detection, druggability assessment and molecular docking
- chem-os — canonical molecular descriptors from RDKit
Comes out
- Predicted structure and pocket map
- Docked poses with an interaction fingerprint
- Descriptor table for the candidate set
Person reviews A structural chemist inspects poses and interaction classifications before any pose is treated as a lead.
How this status was set struct-os exists and its docking skill carries receptor preparation, ligand preparation, a docking runner and an analysis script with tests. Library-scale virtual screening and generative molecular design do not: the virtual-screen-os entry in the product catalogue has no repository under the organisation.
-
06 / 10Partly built Part of this stage has a repository behind it; a named part does not.
Readouts and quality control
Take the raw readout the laboratory produced and decide whether it is fit to analyse.
Goes in
- Sequencing reads or single-cell counts
- Imaging plates
- Sample sheet and batch metadata
Does the work
- BIOos — alignment and sequencing QC
- cell-os — single-cell processing and annotation
- screen-os — library QC, normalisation and batch correction
- statskit — assumption checking and missing-data handling
Comes out
- QC report per sample and per batch
- Normalised matrix
- Explicit list of samples excluded and why
Person reviews The exclusion list is the review artifact: a person signs off on what was dropped before any statistic is computed on the remainder.
How this status was set bioinformatics-os (published as BIOos), cell-os, screen-os and statskit exist and cover sequencing and single-cell readouts. Imaging readouts do not: the bioimage-os directory under the organisation holds a licence, a reproducibility note and a configuration file, with no source and no git history, so microscopy QC is planned rather than available.
-
07 / 10Partly built Part of this stage has a repository behind it; a named part does not.
Hit triage and structure-activity
Separate the hits worth chasing from the artefacts, and look for structure-activity trends across the series.
Goes in
- Ranked hit list
- Assay readouts per compound or per guide
- Candidate structures
Does the work
- screen-os — hit calling, dropout and enrichment scoring, essential-gene reference comparison
- chem-os — molecular descriptors with a provenance record
Comes out
- Triaged hit list with the reason each hit survived
- Descriptor table tied to the input structures
Person reviews Hit calling is a gate, not a report: the call is reviewed against the essential-gene reference and the screen QC before the list is passed on.
How this status was set screen-os carries the hit-calling and scoring skills, so screen triage is real. Chemical structure-activity analysis is not: chem-os holds a descriptor engine and its tests, with no series analysis, no scaffold decomposition and no activity model in the repository.
-
08 / 10Planned No repository does this work today.
Absorption, distribution, metabolism, excretion and toxicity
Predict what a candidate would do in a body, and flag the liabilities that would stop a programme.
Goes in
- Lead structures
- Measured assay activity
- Physicochemical properties
Does the work
Nothing does this work today.
Comes out
Nothing, until the stage is built.
Person reviews Not defined, because the stage is not built. When it is, the review point belongs before any liability call is written into a programme decision.
How this status was set The product catalogue lists admet-os as planned. There is no admet-os directory under the organisation, and no other repository contains absorption, distribution, metabolism, excretion or toxicity prediction. This stage is listed so the journey is complete, not because it runs.
-
09 / 10Planned No repository does this work today.
Translational
Connect a candidate to patients: the model systems, the biomarkers and the cohort evidence that would support a trial.
Goes in
- Lead candidate and its mechanism
- Patient cohort and clinical data
- Biomarker hypotheses
Does the work
Nothing does this work today.
Comes out
Nothing, until the stage is built.
Person reviews Not defined. Cohort and clinical work carries obligations this software does not currently meet, so nothing here should be read as ready for patient data.
How this status was set clinical-os is listed in the product catalogue with no repository behind it. No clinical-os directory exists under the organisation, and no other repository handles cohort or electronic health record analysis. rare-os and clinvar-os cover variant interpretation, which is a different question from translational evidence.
-
10 / 10Planned No repository does this work today.
Regulatory package
Assemble what a regulator would ask for: the method, the data, the decisions and the trail connecting them.
Goes in
- Every artifact produced by the earlier stages
- The review decisions attached to each
Does the work
Nothing does this work today.
Comes out
Nothing, until the stage is built.
Person reviews Not defined. A regulatory package is a legal artifact and is out of scope for this software today.
How this status was set No repository assembles a regulatory submission; the evidence-packaging repositories in the assay journey produce inputs such a package would draw on, nothing more. Worth stating plainly because one name misleads: reg-os is regulatory genomics — enhancers, transcription-factor binding and chromatin accessibility — and has nothing to do with regulatory affairs. No repository under the organisation targets submissions, and this site makes no certification claim of any kind.
What a stage leaves behind
run/ stage: perturbation-and-screening manifest.json inputs, parameters, tool versions library/guides.tsv designed guides with off-target scores qc/guide-level.html counts distribution, representation results/gene-scores.tsv gene-level scores with significance review/gate.json who approved, when, and on what artifact
The products behind the stages
Each stage names the software that does its work. The full catalogue, with the status of every entry, is on the products page.