We present new guidelines for choosing hyperparameters for t-SNE and an evaluation comparing these guidelines to current ones. These guidelines include a proposed empirically optimum guideline derived from a t-SNE hyperparameter grid search over a large collection of data sets. We also introduce a new method to featurize data sets using graph-based metrics called scagnostics; we use these features to train a neural network that predicts optimal t-SNE hyperparameters for the respective data set. This neural network has the potential to simplify the use of t-SNE by removing guesswork about which hyperparameters will produce the best embedding. We evaluate and compare our neural network-derived and empirically optimum hyperparameters to several other t-SNE hyperparameter guidelines from the literature on 68 data sets. The hyperparameters predicted by our neural network yield embeddings with similar accuracy as the best current t-SNE guidelines. Using our empirically optimum hyperparameters is simpler than following previously published guidelines but yields more accurate embeddings, in some cases by a statistically significant margin. We find that the useful ranges for t-SNE hyperparameters are narrower and include smaller values than previously reported in the literature. Importantly, we also quantify the potential for future improvements in this area: using data from a grid search of t-SNE hyperparameters we find that an optimal selection method could improve embedding accuracy by up to two percentage points over the methods examined in this paper.

MOTIVATION: Accurate automatic annotation of protein function relies on both innovative models and robust datasets. Due to their importance in biological processes, the identification of DNA-binding proteins directly from protein sequence has been the focus of many studies. However, the datasets used to train and evaluate these methods have suffered from substantial flaws. We describe some of the weaknesses of the datasets used in previous DNA-binding protein literature and provide several new datasets addressing these problems. We suggest new evaluative benchmark tasks that more realistically assess real-world performance for protein annotation models. We propose a simple new model for the prediction of DNA-binding proteins and compare its performance on the improved datasets to two previously published models. In addition, we provide extensive tests showing how the best models predict across taxa. RESULTS: Our new gradient boosting model, which uses features derived from a published protein language model, outperforms the earlier models. Perhaps surprisingly, so does a baseline nearest neighbor model using BLAST percent identity. We evaluate the sensitivity of these models to perturbations of DNA-binding regions and control regions of protein sequences. The successful data-driven models learn to focus on DNA-binding regions. When predicting across taxa, the best models are highly accurate across species in the same kingdom and can provide some information when predicting across kingdoms. AVAILABILITY AND IMPLEMENTATION: The data and results for this article can be found at https://doi.org/10.5281/zenodo.5153906. The code for this article can be found at https://doi.org/10.5281/zenodo.5153683. The code, data and results can also be found at https://github.com/AZaitzeff/tools_for_dna_binding_proteins.