Sequence & representation
Protein Language Model
Learn numerical representations of residues and proteins from sequences.
Representations and model scores are not experimental function.
METHODS / WORK RECORDS
Original method notes and work snapshots, with their evidence, dates and limits intact. These records do not imply newly completed work.
WUTONG / A NOTEBOOK OF METHODS
Methods in the leaves. Questions meeting between branches.
THE NOTEBOOK IN THE LEAVES
Choose a leaf to explore its method, tools and connections.
Sequence & representation
Learn numerical representations of residues and proteins from sequences.
Representations and model scores are not experimental function.
Structure & interaction
Infer candidate three-dimensional structures from model inputs.
Predicted structures and pLDDT do not validate binding.
Public source entry points · method notes, not claims of execution or mastery.
An existing public prediction record. Explore its structure and confidence; it does not establish binding or experimental efficacy.
View public structure data ↗Structure & interaction
Model possible spatial arrangements of interacting molecules.
Docking poses are hypotheses to test.
Design & optimization
Design amino acid sequences conditioned on a given backbone.
Model compatibility scores do not establish affinity.
Public source entry points · method notes, not claims of execution or mastery.
Design & optimization
Propose candidate backbones using generative models.
Sequence design, evaluation and experiments remain necessary.
Public source entry points · method notes, not claims of execution or mastery.
Design & optimization
Noise and denoise an existing backbone to sample nearby designs.
Diversification does not guarantee improvement.
Public source entry points · method notes, not claims of execution or mastery.
Design & optimization
Propose interface sequence variants while keeping selected structure or positions fixed.
Constrained design still needs independent evaluation and binding assays.
Public source entry points · method notes, not claims of execution or mastery.
Design & optimization
Propose reasoned sequence changes using structure and position constraints; this entry covers constrained sequence design.
A design rationale is not experimental evidence.
Public source entry points · method notes, not claims of execution or mastery.
Physics & energy
Score sequence variants with a model to nominate candidates for testing.
Computational scores and experimental mutation effects are recorded separately.
Public source entry points · method notes, not claims of execution or mastery.
Physics & energy
Simulate molecular motion under a specified physical model.
Trajectories depend on sampling, initial conditions and force fields.
Public source entry points · method notes, not claims of execution or mastery.
Selection & experiment
Measure mutation effects using variant libraries and experimental readouts.
Readouts depend on assay conditions and are not universal affinity measurements.
Selection & experiment
Analyze sequence counts, enrichment and candidate families across selection rounds.
Enrichment also reflects amplification and selection biases; it is not affinity.
New methods earn new leaves. Implementation improvements and validation deepen existing notes. Growth follows evidence.
VIRTUAL R&D / TASK ORGANIZATION SNAPSHOT
Research goals, current stages and next steps.
Evidence date:
Compare changed functions from fixed revisions with registered implementations to reduce duplication.
Evidence date:
Document which code path consumes a baseline in automated checks.
Evidence date:
Propose versioned benchmark definitions and ways to inspect their records.
Evidence date:
Reduce coupling between batch work and persistent-service interfaces while organizing older entry points.
Evidence date:
Propose inspection views and access interfaces for evaluation records.
Evidence date:
Propose layers of checks for software and research workflows.
Evidence date:
Connect an additional design method through a provider interface.
Evidence date:
Define the scope and feasibility questions for a bounded data-catalog check.
Evidence date:
Review storage, backup and recovery changes for scheduled monitoring.
Evidence date:
Change a planning reference to reduce dependence on a particular local location.
Evidence date:
Propose interfaces to preserve and recover request evidence before retrying.
Evidence date:
Propose controlled recovery when a service resource allocation expires.
Evidence date:
Propose explicit stopping and continuation boundaries after publication.
Evidence date:
Propose evidence checks for predictive-method research and its written account.
Evidence date:
Record dependency-provenance questions remaining after workflow changes.
Evidence date:
Clarify applicability, refusal explanations and coverage declarations in check planning.
Evidence date:
Record qualification needs involving execution checks, adoption and repeated-run evidence.
Evidence date:
Track recovery and regeneration questions for missing method assets.
Evidence date:
Track safe recovery after an interrupted first installation.
Evidence date:
Retain daily evidence and integrate model interfaces, publication-catalog compatibility and the complete daily workflow.
Evidence date:
Render the collected structured results directly instead of checking again during rendering.
Evidence date:
Track source integration for this engineering proposal and preserve records needed for later acceptance.
Evidence date:
Track source integration for this engineering proposal and preserve records needed for later acceptance.
Evidence date:
Maintain account switching, launcher bindings, reasoning settings, session-projection consistency and billing display.
Evidence date:
Retain observation provenance and measured scope in operations and disk reports.
Evidence date:
Track source integration for this engineering proposal and preserve records needed for later acceptance.
Evidence date:
Track source integration for this engineering proposal and preserve records needed for later acceptance.
Evidence date:
Track source integration for this engineering proposal and preserve records needed for later acceptance.
Evidence date:
Track source integration for this engineering proposal and preserve records needed for later acceptance.
Evidence date:
Track source integration for this engineering proposal and preserve records needed for later acceptance.
Evidence date:
Maintain third-party image-build arguments, dependency transport and trusted recipe bindings.
Evidence date:
Track source integration for this engineering proposal and preserve records needed for later acceptance.
Papers and methods. Research ↗ Capability catalog ↗
VIRTUAL R&D / TASK ORGANIZATION SNAPSHOT
Research goals, current stages and next steps.
Historical records · awaiting an update.
Evidence date:
Allow an optional state omission in unrelated interactive records while strictly checking the selected record.
Evidence date:
Narrow native process identity checks and retain refusal when evidence is incomplete.
Evidence date:
Record explicit aborts for unsubmitted tasks without fabricating execution handles or completion results.
Evidence date:
Reuse reminder read results and preserve coverage evidence and last-good records for incomplete calendar collection.
Evidence date:
Support bounded pagination and resumption while keeping missing identities unknown.
Evidence date:
Make developer guidance explicit about test maintenance, applicability and unexecuted coverage.
Evidence date:
Quarantine minimal message metadata with replay and conflict checks.
Evidence date:
Remove stale guidance and align tool entry points with maintenance checks.
Evidence date:
Align optimization entry points, configurable roles and usage guidance.
Evidence date:
Connect optimization methods through shared constraints, candidate provenance, feedback and recovery.
Evidence date:
Preserve prior actions and failures when research budgets change.
Evidence date:
Connect capability contracts, execution records and engineering checks so coverage gaps can be identified.
Evidence date:
Avoid constructing unused information for name queries while preserving validation boundaries.
Evidence date:
Resolve shared validation tools correctly for an explicitly selected source scope.
Evidence date:
Improve bounded queries, duplicate checks and operation readback while keeping incomplete results unknown.
Evidence date:
Correct cancellation, approval ordering and receipt validation; preserve uncertain operations without replay.
Evidence date:
Recovery from an interrupted first installation needs explicit ownership rules and an operator decision.
Evidence date:
Develop research changes around contextual use of property prediction methods and supporting evidence.
Evidence date:
Improve test collection and qualification of retained replay records so evidence scope remains traceable.
Evidence date:
Small changes could wait for unrelated tests while independent work blocked each other. Select checks by the changed scope and run independent checks separately.
Evidence date:
A service still doing work could be treated as expired while shared state was being updated. Record state at a consistent point and keep liveness signals independent of slower shared writes.
Evidence date:
Completed results could be rejected at handoff because their format was interpreted differently. Use the actual result format and check provenance and completeness before accepting it.
Evidence date:
Avoid treating a browser environment as ready when it cannot open a page.
Evidence date:
Make task state, evidence, blockers and next actions inspectable together.
Evidence date:
Distinguish a non-object document from missing required fields so schema errors are easier to locate.
Evidence date:
Make design, validation and delivery tools easier to find while keeping their responsibilities clear.
Evidence date:
Preserve evidence through storage or database interruptions and make scheduled monitoring recover or fail visibly.
Evidence date:
Prepare a traceable data collection plan for later property-prediction research.
Evidence date:
Preserve original check obligations and accepted data through tool migrations and interrupted retries.
Evidence date:
Let clients recover original request identity and execution evidence before retrying or resuming.
Evidence date:
Distinguish a completed Beijing workday from a genuinely partial UTC activity snapshot.
Evidence date:
Keep comparison context explicit while showing compact views and preserving access to full records.
Papers and methods. Research ↗ Capability catalog ↗