The group lasso is an extension of the lasso method for variable selection in logistic regression models, where variables are grouped. It ensures invariance under groupwise orthogonal transformations and is statistically consistent even when the number of predictors exceeds the sample size, provided the true model is sparse. The paper introduces efficient algorithms for solving the convex optimization problem in logistic regression, applicable to generalized linear models. A two-stage procedure is proposed to achieve sparser models and improved prediction performance. The group lasso estimator is shown to be statistically consistent in high-dimensional settings. The methods are applied to DNA splice site detection, where the predictor space involves categorical variables. The algorithms are efficient and can handle large-scale problems. The paper also presents a statistical consistency theory for the group lasso in high-dimensional logistic regression. A hybrid method combining group lasso with ridge regression is introduced, which often yields better predictions and variable selection. The group lasso and its variants are applied to DNA motif modeling and splice site detection, showing competitive performance with maximum entropy models. The methods are implemented in an R package and are suitable for high-dimensional data.The group lasso is an extension of the lasso method for variable selection in logistic regression models, where variables are grouped. It ensures invariance under groupwise orthogonal transformations and is statistically consistent even when the number of predictors exceeds the sample size, provided the true model is sparse. The paper introduces efficient algorithms for solving the convex optimization problem in logistic regression, applicable to generalized linear models. A two-stage procedure is proposed to achieve sparser models and improved prediction performance. The group lasso estimator is shown to be statistically consistent in high-dimensional settings. The methods are applied to DNA splice site detection, where the predictor space involves categorical variables. The algorithms are efficient and can handle large-scale problems. The paper also presents a statistical consistency theory for the group lasso in high-dimensional logistic regression. A hybrid method combining group lasso with ridge regression is introduced, which often yields better predictions and variable selection. The group lasso and its variants are applied to DNA motif modeling and splice site detection, showing competitive performance with maximum entropy models. The methods are implemented in an R package and are suitable for high-dimensional data.