Enrichment of fibril-forming sequences predicted in the AmylFrag database of sequences compared with a reference set of sequences. The curves plotted with squares and circles show the fractions of predictions with energies below the threshold energy on the x axis obtained for the AmylFrag and control sets of peptides, respectively. The triangle curve (read off the right y axis) gives the ratio of these two fractions, F_a for AmylFrag and F_b for background, as a measure of enrichment. Energy thresholds <−19.5 do not provide enrichment, whereas an energy threshold of −25.5 provides substantial enrichment.

Schematic representation of the 3D profile method with rosettadesign energy for detecting fibril-forming segments. From the crystal structure of the NNQQNY peptide (Upper Left), a set of near-native templates is generated by translations of one of the two β-sheets relative to the other, along three orthogonal directions, as shown (Center). A sequence of interest (Lower Left) is scanned by sliding a window of six residues and mapping each peptide onto the templates in the ensemble. Each mapping of sequence to template is evaluated energetically with rosettadesign. Finally, a putative prediction is made by taking the best-scoring (lowest energy) fit between peptide and template (Right). The putative prediction is accepted as a prediction if its energy is lower than the threshold energy.

Prediction performance of the 3D profile method with rosettadesign energy for predicting fibril-forming sequences from the AmylHex database, shown as receiver-operator characteristic curves. The percentage of correct predictions is shown as a function of percentage of wrong predictions, as the energy threshold is raised from a very low value (good energetic fit) to a very high (poor energetic fit). The curve with circles shows the fit of hexapeptides to the NNQQNY crystal structure; the curve with squares shows the fit by using the entire near-native ensemble (variations of the NNQQNY structure). The diagonal line shows how a random predictor would perform. The curve plotted with triangles (read off the right y axis) traces the probability that the results at each point on the curve plotted with squares could have been obtained by chance. The two minima of the probability curve are indicated by the black and gray lines.

Application of the 3D profile method with rosettadesign energy for detecting fibril-forming segments of proteins known to form fibrils. (a) Lysozyme. The vertical bars represent the lowest energy template matches for each hexapeptide, with the black bars indicating a known fibril-forming segment of the protein (21). Notice that the hexamer predicted to have the lowest energy is within the known fibril-forming segment. Gray and black horizontal lines indicate the permissive and conservative thresholds, respectively, taken from Fig. 1. (b) Myoglobin. The gray vertical bars represent the lowest energy template matches for segments of the protein where there are no experimental data. The black bars indicate the segment of the protein that forms fibrils in isolation (23), and these include the segments with the lowest energies. Gray and black horizontal lines indicate the permissive and conservative thresholds, respectively. (c) Abeta(1–42). The vertical bars represent the lowest energy template matches for each hexapeptide with those colored gray representing residue positions for which there is no experimental data. Black and gray horizontal lines indicate the permissive and conservative thresholds, respectively. Black vertical bars indicate those segments of the peptide for which there is experimental evidence of fibril formation or ordered β-structure. The experimental evidence for fibril formation of various segments is shown above the plot with the hatched boxes representing NMR/EPR (6, 27, 28), the cross-hatched box representing proline-scanning mutagenesis (29), and the solid lines representing positive fibrillization assays (26, 30). Notice that the lowest energy segment is the C-terminal segment known to be important for fibril formation. (d) Tau. The vertical bars represent the lowest energy predictions for each hexapeptide. Gray bars represent positions where there is no experimental data. Black bars indicate the segment of the protein (PHF43) implicated in aggregation, including the position of the known amyloidogenic peptide VQIVYK (4). Notice that this segment is one of the lowest energy segments. Gray and black horizontal lines indicate the permissive and conservative thresholds, respectively.