Model Limitations

Structure-prediction models can accelerate discovery and help prioritize experiments, but every prediction must still be regarded as the output of a complex model that comes with uncertainty. These systems can produce confident-looking results that are incomplete, context-sensitive, or incorrect.

Predictions should ideally be interpreted in the context of independent corroborating evidence: experimental data, orthogonal computational analyses, literature support, and biological plausibility in the specific assay conditions.

As with large language models, strong capabilities can mask certain deficiencies. Structure models can and do make mistakes, and their errors are not always obvious from visual inspection alone.

Confidence metrics such as pLDDT and PAE can be useful for prioritization, but high confidence scores still do not guarantee that a predicted structure reflects real biological behavior in your specific context.

Best practice is to treat predicted structures as hypotheses to validate, not definitive answers. Use them to guide decisions, then confirm key conclusions with external data before relying on them for critical scientific or business decisions.

Volt is not liable for errors, omissions, or downstream outcomes caused by mistakes in structure-prediction model outputs. By using this platform, you acknowledge that you have read and understood the limitations above and accept responsibility for independent validation before acting on any prediction.