Q1VarPredBio is data-derived tool for predicting functional impact of variants (amino acid substitutions) of the KCNQ1 potassium channel. The underlying model of Q1VarPredBio is a neural network with an input layer of 12 neurons, with the first hidden layer of 32 neurons, and second hidden layer of 12 neurons. The final layer consists of 4 neurons making the phenotype predictions on four functional parameters i.e., peak current density (IKs), voltage of half-maximal channel activation (V1/2), activation time constant (tau_act), and deactivation time constant (tau_deact). This model was trained and tested against 125 functionally validated (in vitro electrophysiological study) KCNQ1 variants and 345 non perturbing mutations(like V110V) with biophysical and evolutionary features. The output of Q1VarPredBio is predicted phenotype with confidence score for four functional parameters. Please note that limitation of Q1VarPredBio is the unavailability of the KCNQ1 secondary structure at residue index 1-103, 394-506 and 568-676.

A high confidence score (out of 100) for a parameter signifies more confidence in the prediction. The confidence score is an equally weighted average of the i) percentage of models supporting the prediction out of an ensemble of 25 independently validated models and ii) the absolute difference between decision boundary and predicted output. Table below gives the criteria for high confidence predictions for four phenotypes:


Instead of using the server, you can also view all mutations at available secondary structure. Wild type residues are based off the canonical 676 residue human KCNQ1 sequence.

If you find this webserver useful in your research, please cite Phul S, Kuenze G, Vanoye CG, Sanders CR, George AL Jr, Meiler J (2022) Predicting the functional impact of KCNQ1 variants with artificial neural networks. PLoS Comput Biol 18(4):e1010038 full text link.