True positive and false positive rate of consensus and alignments based methods. a) True positive and false positive rate of a consensus-based method on all NLSs from our dataset. The false positive rate is shown as the error rate per amino acid residue. The diagonal line depicts a ratio of one true prediction per false prediction per amino acid residue. b) True positive and false positive rate of consensus and alignment based methods on classical NLSs from our dataset. The false positive rate is shown as the error rate per amino acid residue. The diagonal line depicts a ratio of one true prediction per false prediction per amino acid residue.

Lysine and arginine content of characterized nuclear localization signals. Plot of the lysine and arginine content of characterized nuclear localization signals with respect to their length. The plot shows the three 'types' of NLSs present in our study.

True positive and false positive rate of our model. True positive and false positive rate of various methods, including our HMM at various posterior threshold and the Viterbi algorithm on our dataset. The false positive rate is shown as the error rate per amino acid residue. The diagonal line depicts a ratio of one true prediction per false prediction per amino acid residue.

True positive and false positive rate of our model on other species. True positive and false positive rate of various methods, including our HMM at various posterior threshold and the Viterbi algorithm on the PredictNLS dataset. This ROC curve was created by counting overlaps. The false positive rate is shown as the error rate per amino acid residue. The diagonal line depicts a ratio of one true prediction per false prediction per amino acid residue.

Alignment of characterized classical nuclear localization signals. Alignment of the residues thought to contribute to NLS binding to importin-α. The residues aligned on the cNLS major binding site were then used as model for a profile HMM approach using HMMer.

Schematic of our two state and four state HMM. a) The two state HMM models two states which are represented by the 'background', which emits residues with the same frequency as the genome, and by the 'NLS' state, which emits residues with the same frequency as the NLSs from our characterized data. b) The four state HMM models four states which are represented by the 'background', which emits residues with the same frequency as the genome, two 'NLS' states, which emit residues with the same frequency as our characterized NLSs, separated by a 'spacer' state which emits residues with the same frequency as the genome.

Posterior trace of Swi5p and Ste5p for our two HMMs. a) Posterior trace of Swi5p, a characterized bipartite cNLS, using our four state model. Output was generated by NLStradamus and highlighted region shows the region of characterized NLS. Black (i) and blue (iii) lines represent the two patches of basic residues while the pink line (ii) represents the spacer. Green line represents the sum of the three NLS states. Red line is shown as a reference for a threshold of 0.6. b) Posterior trace of Swi5p, a characterized bipartite NLS, using our simple two state model. Output was generated by NLStradamus and highlighted region shows the region of characterized NLS. Horizontal red line depicts the chosen posterior threshold of 0.6. c) Posterior trace of Ste5p, a characterized bipartite importin-β dependent NLS (non-cNLS), using our four state model. Output was generated by NLStradamus and highlighted region shows the region of characterized NLS. Black (i) and blue (iii) lines represent the two patches of basic residues while the pink line (ii) represents the spacer. Green line represents the sum of the three NLS states. Red line is shown as a reference for a threshold of 0.6. d) Posterior trace of Ste5p, a characterized bipartite non-classical NLS, using our simple two state model. Output was generated by NLStradamus and highlighted region shows the region of characterized NLS. Horizontal red line depicts the chosen posterior threshold of 0.6.