Automated peak picking results

[print table]

Spectrum name	Number of peaks				Score (Strong / Weak)				Download [F] - original signal coordinates (folded peaks), [U] - unfolded peak lists
	Expected	Strong	Weak	All	Precision	Recall	F1	Fm	Strong	Weak	All
C13NOESY	-	8467	9641	10000	-	-	-	-	CSV [F], PEAKS [F], LIST [F] CSV [U], PEAKS [U], LIST [U]	CSV [F], PEAKS [F], LIST [F] CSV [U], PEAKS [U], LIST [U]	CSV [F], PEAKS [F], LIST [F]
CBCANH	520	354	456	801	0.878 / 0.844	0.598 / 0.74	0.711 / 0.788	0.724 / 0.79	CSV [F], PEAKS [F], LIST [F] CSV [U], PEAKS [U], LIST [U]	CSV [F], PEAKS [F], LIST [F] CSV [U], PEAKS [U], LIST [U]	CSV [F], PEAKS [F], LIST [F]
CBCAcoNH	288	283	361	741	0.937 / 0.855	0.921 / 1.071	0.929 / 0.951	0.929 / 0.957	CSV [F], PEAKS [F], LIST [F] CSV [U], PEAKS [U], LIST [U]	CSV [F], PEAKS [F], LIST [F] CSV [U], PEAKS [U], LIST [U]	CSV [F], PEAKS [F], LIST [F]
CcoNH_@ALI	451	371	502	1414	0.979 / 0.89	0.805 / 0.991	0.884 / 0.938	0.888 / 0.939	CSV [F], PEAKS [F], LIST [F] CSV [U], PEAKS [U], LIST [U]	CSV [F], PEAKS [F], LIST [F] CSV [U], PEAKS [U], LIST [U]	CSV [F], PEAKS [F], LIST [F]
HCCHTOCSY_@ALI	3443	1243	2078	6960	0.778 / 0.708	0.281 / 0.427	0.413 / 0.533	0.467 / 0.55	CSV [F], PEAKS [F], LIST [F] CSV [U], PEAKS [U], LIST [U]	CSV [F], PEAKS [F], LIST [F] CSV [U], PEAKS [U], LIST [U]	CSV [F], PEAKS [F], LIST [F]
N15HSQC	216	158	171	260	1.026 / 1.003	0.751 / 0.794	0.867 / 0.887	0.878 / 0.893	CSV [F], PEAKS [F], LIST [F] CSV [U], PEAKS [U], LIST [U]	CSV [F], PEAKS [F], LIST [F] CSV [U], PEAKS [U], LIST [U]	CSV [F], PEAKS [F], LIST [F]
N15NOESY	-	3131	4323	8396	-	-	-	-	CSV [F], PEAKS [F], LIST [F] CSV [U], PEAKS [U], LIST [U]	CSV [F], PEAKS [F], LIST [F] CSV [U], PEAKS [U], LIST [U]	CSV [F], PEAKS [F], LIST [F]

Table 1. Results of automated peak picking. Peak lists with low F1 score (< 0.25) are highlighed in red. Frequently, low F1 scores indicate wrong spectrum annotations in the project summary, such as axes transposition (e.g. C-H vs. H-C) or wrong experiment type specification.

Peak picking results compared to 1000+ spectra benchmark

Figure 1. Evaluation of automated peak picking output in reference to a benchmark of 1000+ NMR spectra. Each box plot presents the distribution of the F1 score calculated for all benchmark spectra of the given type. Red dots present F1 scores calculated for the output of this peak picking application call. Relative comparison of F1 scores serves as a quality verification of the peak picking output. Scores above the median (orange line) indicate that the automated peak picking routine was able to process a spectrum more accurately than 50% of the spectra in the benchmark. Low F1 scores might be related to a low signal-to-noise ratio of the spectrum or to a wrong definition of the spectrum type in the project summary.

Data consistency check: spectra referencing

Proper spectra referencing is a major precondition for automated shift assignment and structure calculation. The table below presents results of the CYANA PeakMatch algorithm (Buchner et al., 2013), which has been executed for each pair of spectra that share at least two axes. Each row in the table contains an estimate of the relative shift between target and reference spectrum. The score presented in the last column (Equation 2, Buchner et al., 2013) represents the confidence of shift identification (the higher the better).

The relative shifts listed in the table below should be verified manually and corrected before submitting an automated shift assignment or structure calculation job. They can be applied to the target spectrum using specific tags (e.g. w1:0.025 to increase the chemical shift coordinates in the 1^st dimension of the spectrum and peak list by 0.025 ppm) to define spectrum shifts in the project summary. It is not necessary to repeat automated peak picking after entering such tags in the project summary. The system will apply these corrections automatically.

[print table]

Compared peak lists		Relative shift	Score
Reference spectrum	Target spectrum
CBCANH	N15HSQC	N=0.0, H=0.0	0.72
N15HSQC	N15NOESY	N=0.0, HN=0.0	0.28
CBCAcoNH	N15HSQC	N=0.0, H=0.0	1.03
CcoNH_@ALI	N15HSQC	N=0.0, H=0.0	0.8

¹ Non-zero shifts are significant if greater than 0.2 ppm for ¹³C or ¹⁵N, or 0.015 ppm for ¹H.
² Relative referencing with scores below 0.1 is uncertain.

Table 2. Results of automated spectra referencing. Rows presenting possible referencing errors are highlighted in red. The relative shift is reported for spectra having at least 25 strong cross-peaks (Table 1).

Automated shift assignment results

Figure 1.The diagram illustrates atoms in the protein sequence with color-coded chemical shift information. Dark-blue rectangles represent atoms assigned confidently by the FLYA algorithm, whereas for light-blue ones the assignment is uncertain. The row labeled H^N/H^α shows for each residue H^N on the left and H^α in the center. The N/C^α/C' row shows for each residue the N, C^α, and C' assignments from left to right. The rows βη show the sidechain assignments for the heavy atoms in the center and for the hydrogen atoms to the left and right. In the case of branched side-chains, the corresponding row is split into an upper part for one branch and a lower part for the other branch.

[print table]

Spectrum name	Expected	Assigned expected	Measured	Assigned measured	Ratio	Assigned peak lists
C13NOESY_@FLYA	20234	9321 (46.07%)	9968	5847 (58.66%)	1.6	XEASY Sparky
CBCANH_@FLYA	519	393 (75.72%)	497	350 (70.42%)	1.1	XEASY Sparky
CBCAcoNH_@FLYA	260	230 (88.46%)	394	224 (56.85%)	1.0	XEASY Sparky
CcoNH_@ALI_@FLYA	451	316 (70.07%)	563	301 (53.46%)	1.0	XEASY Sparky
HCCHTOCSY_@ALI_@FLYA	3443	1003 (29.13%)	2660	735 (27.63%)	1.4	XEASY Sparky
N15HSQC_@FLYA	164	149 (90.85%)	175	135 (77.14%)	1.1	XEASY Sparky
N15NOESY_@FLYA	5032	2649 (52.64%)	4918	1727 (35.12%)	1.5	XEASY Sparky
ALL	30103	14061 (46.71%)	19175	9319 (48.6%)	1.5

Table 1. Results of the automated shift assignment. Expected: Number of peaks expected in the spectrum. Assigned expected: Number of expected peaks that are assigned that to a measured (picked) peak. The percentage is relative to the number of expected peaks. Measured: Number of measured (picked) peaks. Assigned measured: Number of measured peaks that have one or more expected peaks assigned to it. The percentage is relative to the number of measured peaks. Ratio: The average number of expected peaks that are assigned to the same measured peak. The average is taken over all measured peaks to which at least on expected peak is assigned. Color coding: Rows corresponding to spectra with low percentage of assigned peaks (<15%) are highlighted in red. The most common factors associated with low percentage of assigned peaks are: (a) wrong spectra axes annotation (e.g. transposition of H-HC axes in C13NOESY, more information), (b) systematic shift between spectra (e.g. HCCHTOCSY shifted in C dimension by -2.7 ppm in reference to C13HSQC), (c) low number of signals that are visible in the spectrum (compared to number of signals that are expected to be present in the spectrum given the protein sequence), (d) inverted axis (inverted ppm scale in one of spectra dimensions).

Automated structure calculation results

Fig. Superposition of two structure proposals (proposal 1, proposal 2) determined by NMRtist (click image to show protein visualization at full scale). Structures are shown in three different orientations. The parts of the structure that are not well-defined are drawn with transparency. The two proposals are calculated with a different number of NOESY cross peaks. Download aligned structure proposals [combined.pdb] . If a "Test run" was executed, NMRtist may output only a single structure proposal.

Property	Proposal 1	Proposal 2
Structure	structure.pdb	structure.pdb
Distance restraints	restraints.upl	restraints.upl
Structure calculation details	cyanatable.txt	cyanatable.txt
Total number of distance restraints	3647	3829
Number of intraresidual restraints (|i-j| = 0)	829	843
Number of sequential restraints (|i-j| = 1)	853	912
Number of medium-range restraints (1 < |i-j| < 5)	1006	1050
Number of long-range restraints (|i-j| ≥ 5)	959	1024
Number of torsion angle restraints	232	232
CYANA target function value	6.64 � 0.26 +�	7.91 � 0.13 +�
Distance restraint violations > 0.2 Å	11 � 2	13 � 2
Maximal distance restraint violation	0.29 � 0.12 +	0.36 � 0.01 +
Angle restraint violations > 5.0°	8 � 0	9 � 1
Maximal angle restraint violation	20.75 � 0.45 �	23.18 � 0.24 �
Residues in most favored Ramachandran plot regions	88.4 %	86.4 %
Residues in additionally allowed regions	11.5 %	13.6 %
Residues in generously allowed regions	0.0 %	0.0 %
Residues in disallowed regions	0.0 %	0.0 %

Table 1. Structure calculation statistics