Reading results
Understand result metrics, interpretation limits, candidate comparisons, and what to share with experimental collaborators.
Computational results include predicted structures, values, and confidence estimates. Understanding what each value measures and its limits helps you compare candidates for experiments.
Computational results and experimental evidence
Keep these five types of evidence separate when interpreting results.
| Type | Meaning | Example |
|---|---|---|
| Computational prediction | Structures or values produced by a model | Predicted complex coordinates |
| Structural confidence | How confident the model is in its prediction | pLDDT, PAE, ipTM |
| Actual binding | Whether two molecules actually bind | Binding observed experimentally |
| Experimental validation | Whether it was measured under defined conditions | Kd measured by SPR |
| Clinical effect | Whether it has an effect in people | Clinical trial results |
The first two are computational outputs; computation alone cannot establish the other three. High confidence means the model is confident, not that the structure has been proven real.
Information on the results page
Metric summary shows values from the run. Output files contain original coordinates, sequences, and tables for download and reuse. Execution evidence records computational conditions such as model weights and input-file hashes. Refer to it when citing or repeating a calculation.
Reading individual metrics
Definitions and ranges may differ by tool. Check the definition in that tool's reference first.
| Metric | Meaning | Typical range | Preferred direction | Do not interpret it as |
|---|---|---|---|---|
| pLDDT | Model confidence in local structure | 0~100 | Higher | pLDDT 90 as "90% accuracy" or "90% binding probability" |
| PAE | Expected positional error in one region after alignment on another | Å | Lower | An experimentally measured error |
| ipTM | Confidence in the arrangement and interfaces of complex components | 0~1 | Higher | Affinity or probability of successful binding |
| TM-score | Global fold similarity between two structures | 0~1 | Higher | Sequence similarity |
| lDDT | Local similarity to a reference structure | 0~1 or 0~100 | Higher | The same measure as prediction confidence (pLDDT) |
| RMSD | Root mean square deviation of matched atomic coordinates | Å | Smaller | A comparison without specifying atom selection and alignment |
| DockQ | Contact and arrangement quality relative to a reference complex | 0~1 | Higher | A value obtainable without a reference structure |
| ΔΔG | Change in a free-energy difference between states or variants | kcal/mol, etc. | Depends on the tool | A value whose sign can be interpreted without checking its convention |
| Tm | Temperature at the midpoint of a thermal transition | °C | Higher under the same conditions | A guarantee of binding, function, and storage stability |
The interpretation of ΔΔG depends on which energy difference it measures. Binding ΔΔG and folding-stability ΔΔG address different questions. Sign conventions and subtraction order vary by tool: negative values can be favorable in one case and positive values in another. Interpret the direction using that tool’s definition and comparison baseline.
Four common misconceptions
"High confidence means binding." ipTM describes confidence in a predicted arrangement. It is neither binding affinity nor the probability of observing binding in an experiment. These quantities measure different things.
"A good docking score means a low Kd." Docking scores support computational search and ranking. The dissociation constant Kd is measured at equilibrium. A good score may help prioritize a candidate for experiments, but must not be converted directly into an affinity value.
"More contacts mean stronger binding." Counts of hydrogen bonds or salt bridges cannot be converted directly into affinity.
"The same metric can be compared across tools." Names may match while definitions and scales differ. Check that definitions match before ranking values from different tools in one table.
Comparing candidates
Compare candidates under the same computational conditions. Different inputs, settings, or model versions make it hard to distinguish candidate effects from condition effects. Compare hashes in the execution evidence to check conditions.
Read global and region-specific values together. A high overall structural confidence can coexist with low confidence in the region of interest.
Failed runs also carry information. Rejected inputs, runtime failures, and low-confidence predictions are different signals. Execution history provides each attempt's status and failure reason.
Citing or repeating results
To cite values in a paper or repeat the calculation months later, retain the inputs and settings used. Execution evidence bundles hashes of model weights and input contents, settings, and the license basis for running that tool.
Identical inputs and settings may still produce different numerical results. Random seeds and execution environments can cause variation. Execution records show what was calculated and under which conditions.
Handing off to an experiment
Send conditions and limitations with the numbers: tool and version, input structure and selected region, metric definitions, and the fact that the values are predictions. Computation helps narrow experimental candidates; it does not replace experiments.
Importing experimental results
Upload the laboratory's CSV under Experimental measurements on the results page and link it to candidates. Measurements are stored separately from predictions and shown in measured-value columns after prediction columns. Replicates remain individual records rather than being averaged.
There are two ways to upload CSV files.
| Method | When to use it | What to do on upload |
|---|---|---|
| The laboratory's original table | An Excel report saved as CSV | Select the candidate-ID column and specify the metric, unit, and conditions for each measurement column. |
| Nadoo Bio standard CSV | Results requested in the standard format | Select the file; metrics, units, and conditions are already included. |
Before requesting results, download a blank template with Get standard template on the upload screen and send it to the laboratory. The returned file can then be uploaded without reformatting.
Standard CSV rules
Write one measurement per row. Two measurements of the same candidate use two rows; KD and Tm measurements also use separate rows.
| Column | Required | What to enter |
|---|---|---|
| candidate_id | Required | Candidate ID from the Nadoo Bio request sheet. This links the measurement to the candidate in the result file. |
| assay | Required | Assay type: binding_screening (binding screening), binding_kinetics (binding kinetics), thermostability (thermal stability), expression, enzyme_activity (enzyme activity). |
| method | Required | Measurement method, e.g. SPR, BLI, nanoDSF. |
| metric | Required | Metric: KD, kon, koff, Tm, Tonset, Tagg, yield, concentration, binding_call, binding_strength, expression_call. |
| provider | Required | Measurement provider |
| value, unit | Numeric metrics | Numbers only in value; put the unit in unit. |
| value_text | Categorical metrics | The binding_call, binding_strength, or expression_call classification, e.g. true or strong. |
| qualifier | Optional | Leave blank for an exact value. < and > mean less than and greater than. ND means not detected; leave the value empty. |
| replicate | Recommended | Replicate number 1, 2, 3…; use mean for a provider-calculated average. |
| target | Required for binding measurements | Binding partner molecule |
| temperature_c, buffer, condition_note | Recommended | Temperature (°C), buffer, and other conditions. |
| uncertainty, uncertainty_type | Optional | Uncertainty and its type (sd, sem, ci95_low, ci95_high, log10_sd). Provide the type whenever a value is given. |
| provider_sample_id, measured_at, source_file, notes | Optional | Provider sample ID, measurement date (format: 2026-10-20), original filename, and notes. |
Each metric has a defined set of units. Unsupported units are flagged before upload.
| Metric | Unit |
|---|---|
| KD, concentration | M, mM, uM, µM, nM, pM (comparisons use a separately converted value in M). |
| kon | 1/(M*s) |
| koff | 1/s |
| Tm, Tonset, Tagg | degC or °C |
| yield | mg/mL, mg/L, ug/mL |
Detection limits and nondetection retain their meaning. <5 is not converted to 5, and ND is not converted to 0. Values containing units in the same cell ("12 nM") and duplicate replicate numbers under identical conditions are rejected. Do not put predictions such as ipTM, pLDDT, or docking scores in this file.
The following fictional values demonstrate the format; they are not experimental results.
candidate_id,assay,method,metric,qualifier,value,value_text,unit,replicate,target,temperature_c,provider
cand-01,binding_kinetics,BLI,KD,,12.5,,nM,1,Target A,25,Example laboratory
cand-01,binding_kinetics,BLI,KD,,13.1,,nM,2,Target A,25,Example laboratory
cand-01,binding_kinetics,BLI,binding_call,,,true,,1,Target A,25,Example laboratory
cand-02,binding_kinetics,BLI,KD,ND,,,nM,1,Target A,25,Example laboratory
cand-02,thermostability,nanoDSF,Tm,,61.4,,degC,1,,,Example laboratorySample request to the laboratory
Please return measurements in the attached CSV template, with one measurement per row. Copy candidate IDs from the request sheet into candidate_id. Enter numbers only in the value field and units in unit. For nondetection, set qualifier to ND and leave the value blank. Please report each replicate in its own row rather than only an average; if also reporting a mean, set replicate to mean. Please include the original instrument reports (PDF or Excel).
Linking to a specific structure region
Create a link to specific chains and residues in a PDB/mmCIF file. Opening it selects that region in the 3D viewer.
https://nadoo.bio/app/projects/<project-id>/structures/<artifact-id>?sel=A:45-55,A:60,B:10A- Numbers are author residue numbers from the structure file. Negative values (
A:-2) and insertion codes (B:10A) are supported. They may differ from positions in the sequence string. - Positions without coordinates are not selected; they are listed separately as "Positions absent from this structure." If a range endpoint lacks coordinates, only existing residues within the range are selected.
- For a file with multiple models, choose a model on screen or add
model=to the link. - Recipients need permission to view the file in its project. If not signed in, they return to the same link after signing in.
- AI tools create the same link by passing
kind: "structure"andselectiontocreate_view_link.
Preparing Boltz batches
Prepare multiple plans from single protein sequences in CSV/FASTA, and check inputs, setting combinations, and submission eligibility.
Independent K-Fold benchmark record
Measurements from running the public complex-structure model K-Fold: confidence, runtime, and interpretation limits for five protein targets and one ternary complex.