What if we could change an insect’s decision to land?

Given molecular structure, predict the probability of the target insect landing in a standardized laboratory assay to within 10 percentage points of the measured landing rate for every previously untested compound in a predefined test set.

Specifically, predict the probability that an individual insect lands on a designated target at least once during a fixed observation period. Before evaluation, define the insect species, sex, life stage, physiological state, target surface, compound exposure, and environmental conditions. These conditions are held consistent across the evaluation, with untreated and reference controls included.

The test set must be selected independently of the model developer and contain compounds whose behavioral results were not used to train, tune, or select the model. Molecular structures must specify stereochemistry where applicable. The evaluation should include unfamiliar chemical families to test whether the model can generalize beyond closely related compounds already represented in its training data.

Predictions must be recorded before experimental outcomes are revealed. Each prediction is compared with the measured proportion of insects that land during the observation period. For example, a measured landing rate of 60% requires a prediction between 50% and 70%.

Every compound must meet the 10-percentage-point tolerance. Errors cannot be averaged across compounds to conceal a failed prediction. If multiple exposure conditions are included, each compound must meet the tolerance in each specified condition.

Experiments must include enough independent observations to estimate landing rates with precision appropriate to the tolerance. The sampling plan, treatment of inconclusive measurements, and pass-or-fail calculation must be fixed before evaluation. The result must then be reproduced in an independent laboratory using the same locked model and assay protocol.

Given molecular structure, predict the probability of the target insect landing in field trials under defined exposure and environmental conditions, to within 10 percentage points of the measured landing rate for every previously untested compound in a predefined test set.

Specifically, predict landing on a designated target during a fixed observation period in the intended field setting. Before evaluation, define the target insect, host or crop, formulation, delivery method, exposure conditions, observation area, and measurement method. Record relevant environmental conditions, including temperature, humidity, and airflow, using a protocol established in advance.

The trials must provide a defensible estimate of landing probability. Define which insects had an opportunity to land and how individual landing outcomes are recorded. Repeated landings by the same insect must not be treated as independent insects, and landing counts without an appropriate denominator cannot substitute for a probability.

The model, permitted inputs, test compounds, field settings, and evaluation rules must be fixed before behavioral outcomes are revealed. The compounds’ own behavioral results must not have been used to train or adjust the model. If the same compounds are evaluated in both challenges, field predictions must be locked before their laboratory results are disclosed.

Every compound must meet the 10-percentage-point tolerance in every predefined field condition. Performance must be evaluated separately across the specified locations, exposure conditions, and observation periods. Accurate predictions in one setting cannot compensate for inaccurate predictions elsewhere.

The evaluation must include independent field trials and sufficient replication to distinguish model error from sampling uncertainty. Success establishes that the model predicts landing under the tested field conditions; the conditions covered by that result must be stated explicitly.