A local comparison across conformations
The method follows loop-like features in the protein’s Cα geometry and compares their persistence across static or generated conformations. Candidate regions are checked against background and null controls. A strong local difference suggests where to inspect the structure; it does not measure the fraction of time a pocket is open.
Fast alternatives already exist. PocketMiner predicts cryptic-site propensity from a single structure; CrypToth combines persistent homology with mixed-solvent dynamics. BioTwin investigates local topological differences and the scientific questions that follow.
| Approach | Signal used | BioTwin comparison | Question for the next step |
|---|---|---|---|
| Learned cryptic-site prediction | PocketMiner learns structural patterns associated with cryptic pockets. | BioTwin’s local Cα method compares persistence features across supplied or generated conformations. | Does this target’s structural change nominate a region worth investigating? |
| Topology with solvent sampling | CrypToth uses mixed-solvent MD and persistent homology to analyze pocket behavior. | BioTwin’s examined route uses local Cα structural comparisons and matched/null controls. | Is the signal relevant to a possible opening, or explained by a structural control? |
| From a region to a mechanism | A nominated pocket needs subsequent structural and chemical assessment. | Related BioTwin methods separately examine all-atom access, joint ensemble ligandability and capture kinetics. | Can a molecule reach a relevant state and support useful chemistry there? The combined route needs target-specific validation. |
The approaches differ in inputs and outputs. No matched accuracy or cost advantage is claimed. The earlier study and current development are distinguished below.
Develop the opening into a testable intervention.
A promising region becomes useful when you can explain how an intervention might reach it, change something important and survive a fair test. BioTwin has methods for each of those questions, alongside records that keep the evidence connected as the investigation develops.
This is a research pathway through existing methods, assembled for the target and question. Execution support, input compatibility and validation differ by step; the full sequence is not a universally validated automatic discovery loop.
- 1. Find a place to investigate · Compare structural states and local pocket signals, including matched controls. A nomination identifies a question worth pursuing.
- 2. Make the molecular proposal concrete · Examine atom-based fit and entry, attachment geometry and chemical reach. Keep the source structure and uncertainty that could change the answer.
- 3. Consider another route to the same effect · Explore whether a different chemical handle, an allosteric perturbation or a recruitment mechanism could satisfy the biological objective. Each alternative brings requirements to check.
- 4. Keep the reasoning inspectable · Record assumptions, source relationships and scoped outcomes. Distinguish a detected site, no site found within a defined search, and an unresolved result. Agreement between outputs with a shared source is not independent confirmation.
- 5. Specify the result that would change the decision · Use supported next-experiment and assay-planning methods to turn an uncertainty into a testable prediction, appropriate controls and proposed measurement conditions.
- 6. Revise the hypothesis as evidence arrives · In supported native program workflows, recorded outcomes and human guidance can inform the next assessment. Continue a promising route, change the mechanism, or retain a rejected branch with the reason it was set aside.
Keep the objective while changing the approach.
For example, a candidate region may pass a geometric check but lack suitable cysteine chemistry. That creates a new question: could another attachment residue or a different control mechanism produce the same intended effect? The next investigation should distinguish those possibilities, not simply repeat the original score.
- Geometry and access Examine fit and entry
- The biological objective Connect the site to a protein-state objective
- Connected evidence See how evidence supports iteration
Make the first search practical on ordinary hardware.
Local structural comparisons provide a route to candidate regions that can be investigated within a desktop or laptop workflow. Start with a focused structural question, then use more detailed fit, chemistry or simulation work where it can change the decision. This staged approach has a different purpose from simulating complete atomistic dynamics.
- Plan the analysis around your machine · Target size, conformer count and the analysis settings determine the workload. Research on your computer
The controls changed the interpretation
The earlier apo-only study tested a specific C-alpha topology detector and did not establish a general discovery advantage. BioTwin has since added atom-based geometry, separate fit and entry checks, and evidence-aware result handling. Those additions strengthen candidate assessment; their combined blind-discovery advantage still needs a matched evaluation.
Selected known-site and two-structure studies localized useful regions, and a structural null explained much of one apparent mutation-specific signal. Those findings make matched structures and negative controls essential. Side-chain pockets may also be missed by a Cα representation.
Questions behind a useful pocket result
- Does an empty search mean the protein has no useful site? · It means no site was found under the stated representation, structural inputs and search conditions. A cavity, a surface groove and a channel ask different geometric questions. A method's result should stay within the type of site it examined.
- Does a generated opening tell us how often it occurs? · A generated conformation supplies a structural hypothesis. Estimating its frequency or lifetime requires suitable sampling or experimental evidence. BioTwin has an existing all-atom molecular-dynamics source, but using that source does not by itself establish converged populations for a particular target.
- What should a fair evaluation compare? · The same permitted starting information, target structures, site-selection rules and readout. Candidate discovery, assessment of a known site and prediction of binding should be evaluated as different tasks.
- What can the investigation deliver before laboratory testing? · A more specific intervention hypothesis, its supporting calculations, the assumptions that remain open, and a proposed test that could support or rule out the mechanism. Computational results can justify a well-defined experiment; biological engagement and function still require appropriate measurements.