Breaking boundaries: TINTO in POKY for computer vision-based NMR walking strategies

Nuclear magnetic resonance is a crucial technique for studying biological complexes, as it provides precise structural and dynamic information at the atomic level. However, the process of assigning resonances can be time-consuming and challenging, particularly in cases where peaks overlap, or the data quality is poor. In this paper, we present TINTO (Two and three-dimensional Imaging for NMR sTrip Operation via CV/ML), an advanced semiautomatic toolset for NMR resonance assignment. TINTO comprises two separate tools, each tailored for either two-dimensional or three-dimensional imaging. The toolset utilizes a computer-vision approach and a machine learning approach, specifically structural similarity index and principal components analysis, to perform visual similarity searches of resonances and quickly locate similar strips, and in that way overcome the challenges associated with peak overlap without requiring peak picking. Our tool offers a user-friendly interface and has the potential to enhance the efficiency and accuracy of NMR resonance assignment, particularly in complex cases. This advancement holds promising implications for furthering our understanding of biological systems at the molecular level. TINTO is pre-installed in the POKY suite, which is available at https://poky.clas.ucdenver.edu.


Introduction
Nuclear Magnetic Resonance (NMR) is a powerful technique for studying biological complexes, but the resonance assignment process can be time-consuming.To automate this process, various tools have been developed, including AUTOASSIGN (Zimmerman et al. 1997), GARANT (Bartels et al. 1997), I-PINE (Lee et al. 2019), ssPINE (Dwarasala et al. 2022), APSY-NMR (Dutta et al. 2015), BASARA (Bishop et al. 2023), and others, and they accept peak lists as inputs.Due to making peak lists using these automated assignment tools is time-consuming, additional tools have been developed to automate peak picking.These tools use various techniques, such as contour diagrams [CAPP (Garrett et al. 1991)], Bayesian methods (Cheng et al. 2014), neural network (Nawrocka et al. 2022), AUTOPSY (Koradi et al. 1998), APES and iPick in POKY and NMRFAM-SPARKY (Shin et al. 2008;Rahimi et al. 2021;Lee et al. 2015Lee et al. , 2021)), and more, to identify peak resonances in the spectrum, but incomplete or incorrect assignments may still occur due to the poorly detected peaks.Indeed, peak picking remains a challenge in complex biomolecules.
While these tools can be highly effective in ideal conditions with high-quality data, incomplete or incorrect assignments may occur when the data quality is poor or when the resonances overlap.To address this issue, we introduce a new method, called TINTO (Two and three-dimensional Imaging for NMR sTrip Operation via CV/ML), a new assignment assistant by CV-based strip matching tool that does not rely on cross-peaks for sequential walking on the spectrum.TINTO uses computer-vision and machine learning (unsupervised learning) approaches, including Structural Similarity Index (SSIM) (Wang et al. 2004) and Principal Component Analysis (PCA) (Jolliffe and Cadima 2016), to perform visual similarity searches of resonances and quickly locate similar strips (usually represented as i + 1 and i − 1).
Although the SSIM has traditionally been used for image analysis and has not been applied in NMR data much, Principal Component Analysis (PCA) has been commonly used for comparing NMR data (Xu and Doren 2018), by calculating scores plot of PCs (Principal Components).PCs are new variables obtained by finding linear combinations of the original variables that preserve as much variability as possible in the dataset.PC1 and PC2 specifically refer to the first and second principal components, respectively.These components are derived from an eigenvalue/eigenvector problem using the covariance or correlation matrix of the data.PCs are uncorrelated, representing independent directions of variability, and their variance indicates their importance in capturing overall data variability.The first few principal components are typically of primary interest as they explain the majority of the dataset's variability (Rahimi et al. 2021).
Inside TINTO, we calculate each PC1 or PC2 value for every NMR strip, which is treated as an image, and we use the similarity of their PC1 values to determine the similarity between strips.Our experimental results demonstrate that using PC1 values provides more accurate results than using PC2 or PC3 values.Our approach represents a novel application of PCA and SSIM for NMR assignment, where NMR strips are considered as images.
Figure 1 provides examples of different cases where Peak-based Matching (PM) does not work, but Computer Vision (CV) does.In an ideal case (Fig. 1a) with good peak dispersion, both PM and CV work well.However, when there are overlapped signals (Fig. 1b-d), PM cannot find the right strips due to the low accuracy of the peak picking.To address this limitation, we have investigated the possibility of managing NMR spectra as digital images and employed computer-vision and machine learning (ML) algorithms.In addition, TINTO is a highly customizable and versatile tool that allows easy data visualization and manipulation.Overall, TINTO shows great potential for addressing the challenges associated with NMR assignments in the presence of signal overlap and poor data quality.

Methods
TINTO is primarily written in Python and employs two key techniques: SSIM and PCA, which are implemented using the Scikit-image (skimage) and Scikit-learn (sklearn) packages, respectively.To accommodate two different multidimensional NMR commonly used for proteins, we have developed two versions of TINTO.The stand-alone version of TINTO provides a graphical user interface that is specifically designed for 2D NMR data, while the integrated strip plot version of TINTO, shown as [?] Ssim and [?] Pca buttons, where [?] means 'search', allows users to find the best strip matching from loaded strips on the strip plot.Figure 2 c Shows the case where there are many peaks in the same HN chemical shift, corresponding to multiple spin systems, which make difficult to associate the right peak to the right amino acid residue and at the same time the chemical shift information of the i + 1, note that in the ideal case there is only one spin system per HN chemical shift shows the graphical user for NMR computer vision assignment of 2D and 3D data.Both versions utilize SSIM and PCA to identify the most similar data and offer a range of preprocessing options, including pareto, unit, raw, absolute, minmax and above-noise, all the possible options are going to be explained in the section 'User customizable options of TINTO'.Appropriate preprocessing steps can significantly enhance the results, and TINTO provides these options to ensure the best possible output.

Test sets used for TINTO
The performance of both versions was evaluated by two distinct protocols, as described in Supplementary material SA, SB and SC.We tested TINTO 2D using 2D TOCSY and NOESY spectra (Fig. S1) and TINTO 3D using 2D 15 N-HSQC spectrum and 3D HNCACB and CBCA(CO)NH spectra (Fig. S2).Our protocol has been extensively tested and validated, providing users with a reliable tool for their research.Additionally, users have the flexibility to customize the protocol to meet their specific needs and objectives.
We used three proteins as test sets, consisting of two experimentally acquired data sets and one simulated data set (Table 1).The simulated data set was generated using POKY's SIM-PROC module to simulate backbone 3D experiments of ubiquitin (two-letter-code "me").The other data sets were obtained from our published works (López-Giraldo et al. 2020;Lee et al. 2021).

User customizable options of TINTO
TINTO provides users with a high level of customization, enabling them to adjust various settings to meet their specific needs (Fig. 2b, d).For example, in the 3D strip plot version, users can set the number of best strips they want to see, which determines the number of similar strips displayed in the plot (default value is 10).Additionally, users can choose from different masking options to subtract peaks with the same chemical shift of two spectra and highlight the remaining peaks, making it easier to identify the next residue.For instance, a theoretical spectrum made by HNCACB minus CBCA(CO)NH will exhibit CA i and CB i by eliminating CA i−1 and CB i−1 .By that way, finding a matching strip from CBCA(CO)NH becomes more accurate.Choosing the appropriate masking option should be based on the spectrum combination being analyzed, the sequence location in the strip plot, and the selected strip before clicking on SSIM or PCA. Figure S3 illustrates the different masking options and how they can be used.
Because of the intensity discrepancy underlying the data, TINTO provides a way to customize data scaling in the option menu (Fig. 2b, d).The Pareto option divides all intensity values by the square root of the standard deviation of all values.The Unit option divides all intensity values by the standard deviation of all data.Raw option means that the data is not processed by either Pareto or Unit.The Absolute option returns the absolute value of each element to avoid negative intensities that can complicate data comparison, especially in spectra that have both negative and positive signals.The Minmax option scales the intensity values between 0 and 1.Additionally, the software includes an above-noise option that automatically calculates the spectrum noise and sets all intensity signals below this value to zero.Finally, users can choose to exclude the same spectrum, in this way the original provided spectrum is not going to be part of the results list.
Exclusively to TINTO 2D, users have access to additional options (Fig. 2b), such as Strip Buffer, Searching Method, Exclude Water, and Minimum Separation.The Strip Buffer option refers to the size of the strips being compared.For instance, if the value is set to 0, it will compare the intensity data of each data point, as it will cover a range of [0, 1), corresponding to one data point.On the other hand, if the value is 2, it will use 5 data points, as it will cover a range of [− 2, 3), equivalent to the points − 2, − 1, 0, 1 and 2. The Searching Method includes SSIM and PCA, which is going to determine the algorithm to use when searching similar strips.The Exclude Water option is particularly helpful when working with 2D data collected with water suppression, where water interference may be present.The user can enter the chemical shift of the water signal in the spectrum, and the algorithm will avoid the data present at that chemical shift.The Minimum Separation is going to determine the minimal chemical shift difference between the results.These options provide users with the flexibility to customize the analysis process further and achieve more accurate and reliable results.
Moreover, TINTO 2D includes a feature that allows users to select the View reference of the strip to use as a pattern reference for finding similar strips, as well as the spectrum where it will be located.For example, in the View spectrum, the user can select the 2D TOCSY located at the i amino acid chemical shift, and then select the NOESY as the spectrum finder to identify similar patterns for the i + 1 amino acid.The results will show the most similar patterns, including the i + 1, i − 1, and any spatially close residue.Based on our findings, we have established the default settings for TINTO that have yielded successful results.Figures 2b, d showcase these default options for TINTO 2D and TINTO 3D, respectively.For TINTO 2D, these options include a 0-strip buffer, Pareto optimization, SSIM, and a minimum separation of 0.03 ppm.For TINTO 3D, the default settings comprise 10 best strips, − 1 masking option, Pareto optimization, absolute scaling, minmax normalization, and exclusion of the same spectrum.

TINTO performance on 2D TOCSY/NOESY
The performance of TINTO was evaluated by using 2D TOCSY and 2D NOESY spectra of Cl13. Figure 3 illustrates a sample output from its performance evaluation at residue number 25.Because TINTO uses currently displayed data as a reference, we have tested different view extents.
Our results showed that limiting the spectral window of the Spectrum finder to the H N -lateral chain region performs better in general.This strategy minimizes interferences such as water suppression signals or diagonal peaks, resulting in more accurate results.This protocol is particularly useful for identifying i + 1 residues.However, for users interested in identifying i − 1 or residues that are spatially close, we recommend utilizing the NOESY spectrum as a View reference and Spectrum finder to follow the outlined steps.
The overall results obtained from testing our TINTO 2D on Cl13 are summarized in Fig. 4, together with the SSIM and PCA results with two different variations, as presented in the supplementary material (Table S1). Figure 4 depicts a heatmap of the rank values for each amino acid residue, utilizing four different variations (refer to Table 2 for details).
Here, "rank" denotes the position of the i + 1 chemical shift in the list of results provided by TINTO 2D.Lower rank values indicate better results, indicating a faster and more accurate identification of the i + 1 residue.The "ssim_default" variation demonstrates the best performance, with a majority of residues being found within the first rank 10, represented in dark green.
Also exemplified on figure S4, for the first variation, the results indicate that 89.95% of the residues i + 1 were found within the top 10 strips, while 8.62% of the results were in the strips between 10 and 40, and the remaining 1.72% were located in strips beyond the top 40.We observed a high likelihood of finding the i + 1 residue within the first 10 strips when using the default options, with a greater frequency of occurrence within the first 5 results.
In the second variation of our protocol (SSIM_MIN-MAX, Table 2), we included the minmax option in addition to the default options for running SSIM.This led to 48.27% of the i + 1 residues being located within the top 10 strips, 36.2% located in between the results 10 and 40, with the remaining 15.52% located in strips beyond the top 40.Notably, we observed a significant increase in the SSIM value after applying minmax to normalize all the values between 1 and 0.
All the results we collected assumed a deviation of ± 0.1 ppm from the correct value, this deviation Fig. 3 Test of residue 25 (i) using TINTO for 2D data.Residue 26 (i + 1) chemical shift is 7.813 ppm.We highlight in red the three best candidates found by TINTO, as well as on the right side the corresponding strips.All the results show some similarity with the original peak strip corresponds to a view window of 0.2 ppm, this is because we are considering the data as images.Due to this approach, the peaks may be located in either the left or right side of the view window.However, we also noted that when there are two or more sequential spin systems in the same H N chemical shift, the results are dependent on the deviation entry.For example, if the chemical shift for residue i is 8.1 ppm and the chemical shift for residue i + 1 is 8.13 ppm, but the deviation entry is 0.04, the program may not be able to find i + 1. Conversely, if the results deviation is too low, if the same peak has a diameter lower than this value, the first result could be itself.Additionally, the presence of different spin systems in the same chemical shift can also make it more difficult to locate i + 1 for a specific residue in that chemical shift.
We also tested a single residue and evaluated its performance with various preprocessing options.Figure S5 displays strips representing the results of the default, minmax, absolute, and above-noise preprocessing options, along with the first ten outcomes for each case.The chemical shift of the i + 1 residue is 7.813 ppm, and we observed that the first outcome in all cases is near this value, indicating that the correct residues have been identified in each case.
We also assessed the effectiveness of PCA as an algorithm for identifying similar strips.In Fig. 4, we present a summary of our findings from testing the protocol using PCA with two variations, which are detailed in the supplementary material (Table S2).The table includes ΔPC1 values, which indicate the discrepancy between the PC1 value of the chosen peak and the PC1 values of other strips.We opted for PC1 as the preferred measure for this application based on our observations, as PC2 did not yield satisfactory results in locating the i + 1 residues.From Fig. 4, it Fig. 4 TINTO 2D performance on Cl13 NMR data using different variations.The tested variations include: SSIM with default settings, SSIM with the minmax option, PCA with default settings, and PCA with the minmax option.The colors in the heatmap represent the rank of the i + 1 value in the list of results, with lower values indicating a more efficient method.The ideal rank is 1 (green), while the maximum rank is 40 (red) is evident that PCA does not yield results as promising as SSIM_DEFAULT, as it tends to show a rank distribution more skewed towards higher values.
For a more detailed breakdown of the results, Figure S4 illustrates that when using the default options (PCA_ DEFAULT, Table 2), 50% of the i + 1 residues were found within the first 10 results, 34.48% were located between the 10th and 40th results, and the remaining 15.52% were found beyond the top 40 results.With the minmax preprocessing (PCA_MINMAX, Table 2), we observed a slight improvement compared to the default options, with 56.89% of the i + 1 residues found in the first 10 results, 31.03% between the 10th and 40th results, and the other 12.06% located beyond the top 40 results.
Although both SSIM and PCA demonstrated effectiveness in identifying similar strips, our results indicate that SSIM outperforms PCA for this specific application.

TINTO performance on triple resonance experiments
To evaluate TINTO's performance on triple resonance experiments for backbone assignments, we used spectral data from Nsp7 and Ubiquitin as described in Table 1.Specifically, we used 2D 1 H, 15 N-HSQC, 3D HNCACB and 3D CBCA(CO)NH spectra as inputs.We employed TINTO in the strip plot tool (two-letter-code "sp") to find the strip from the following residue and evaluate its "sequence walking" performance.Figure 3 shows a sample output and visualization of the program's performance.Our results indicate that, similar to the 2D version of TINTO, the size of the strips is critical in obtaining accurate outcomes.Therefore, we recommend adjusting the strip size to focus only on peaks to search before using the TINTO buttons for SSIM and PCA.TINTO proved effective in assigning resonances even in challenging cases where good peak dispersion was lacking, as shown in Fig. 5.
Figure 6 summarizes the TINTO 3D performance results obtained from the tests conducted on Ubiquitin and Nsp7.The colors in the figure correspond to the rank values, representing the position of the i + 1 residues in the strips results.Lower rank values indicate better results, and as depicted in Fig. 6, darker shades of green color indicate higher efficiency in the utilized method.The results show that TINTO was able to accurately find the location of i + 1 residues in both proteins in most cases.However, it is evident from the figure that SSIM significantly outperformed PCA, showing a much better performance overall.For a more detailed view of the results, please refer to the supplementary material (Table S3, S4).
Figure S6 shows, for Ubiquitin, 97.1% of the 'i + 1' residues were located in the first 10 strips (%Rank ≤ 10) using the SSIM with default settings.A higher frequency of i + 1 residues was found in the first strip.Only 2.9% of the i + 1 residues were found between strips 10 and 40 (%Rank > 10 < 40), while 0% were found beyond strip 40 (%Rank ≥ 40).We noted a significant difference in results when testing with wider versus thinner strips.Thinner strips proved to be a better input for TINTO, as they reduced the risk of interference from other peaks belonging to different spin systems.Wider strips increased the probability of finding i + 1 residues in later strips, possibly due to the presence of additional peaks.
For the best results, it is ideal to use a strip view containing only the peaks of the amino acid under study.This would allow TINTO to find similar images with the necessary information and improve the precision of the results.Figure 5 shows an example of the ideal width strip, where the strips are similar in width to the signals.For instance, if the horizontal diameter of the signals is 0.5 ppm, the ideal strip width would be ~ 0.5 ppm. Figure 2c illustrates an example of wider strips, where multiple spin systems can be observed in strips 3-7.When the user opens the strip plot, the default view is similar to Fig. 2c.However, the user can adjust the size of the strips using the >-< button to decrease the strip width.For Nsp7, 83.54% of the i + 1 residues were found within the top 10 strips, with 11.39% of the results between strips 10 and 40, and 5.06% located beyond strip 40.The rank 10 results for Nsp7 show some differences compared to Ubiquitin, mainly due to Nsp7 being experimental data and a larger protein by approximately 9 amino acids.Nonetheless, the results are still impressive.
Additionally, we conducted a comparative analysis between TINTO, a novel sequential walking tool, and the conventional assignment tool, I-PINE, using Nsp7 data.Due to I-PINE's requirement for peak picking, we utilized APES, an automatic peak picking tool, as a preprocessing step.The APES/I-PINE results are presented in Fig. 6, with white indicating successful assignments by I-PINE and black representing unsuccessful assignments.Remarkably, TINTO could find a strip from the next residue in most cases where I-PINE failed (black bar in the APES/I-PINE column).We acknowledge that I-PINE is a valuable automatic assignment tool, but its heavy reliance on peak data quality renders it less effective in situations with significant peak overlaps.To address this limitation, we propose TINTO as a complementary tool, offering an alternative approach that does not depend on peak information.TINTO's introduction into the assignment process aims to accelerate and enhance the overall analysis.
Figure 7 displays the assignment results for residues 3, 4, and 5.In these instances, I-PINE struggled to identify the correct assignments due to signal overlap (Fig. 7d) and the presence of multiple signals on the same HN chemical shift (Fig. 7f).Conversely, TINTO successfully achieved the correct sequential assignments in all these cases.
To evaluate the performance of TINTO on a single residue, we tested various preprocessing options and analyzed their impact on the program's output.Figure 8 illustrates the results of our tests, which included the default preprocessing Fig. 5 TINTO analysis of residue V13 (i) in Nsp7 protein.The residue L14 (i + 1) (marked with a star) is in the strip results (rank 3 followed by L60 and Q35).The box (left) highlights the presence of highly overlapped signals in the region, which can make it chal-lenging to identify and assign resonance.TINTO finds similar looking strips based on computer vision estimator, SSIM, without relying on conventional peak-based matching that requires peaks picked and deconvoluted option, as well as minmax, absolute, and above-noise options.Our analysis revealed that both the default and absolute preprocessing options produced the desired i + 1 result in rank 1, while the above-noise option placed the i + 1 result in rank 4, with the first three strips showing good similarity to the original.However, when we used the minmax preprocessing option, the i + 1 result was in rank 34, indicating that scaling the intensity values between 0 and 1 is not optimal for TINTO's performance.
We evaluated the effectiveness of PCA algorithm for identifying similar strips using 3D data, following a similar protocol to TINTO 2D. Figure 6 presents a summary of our findings from testing the protocol with Ubiquitin and Nsp7 data, as detailed in Table S3 and S4 of the supplementary material.The table shows the Rank values, which correspond to the strip position of the i + 1 residue.
Our results indicate that using the PCA algorithm does not yield satisfactory results in this application, particularly with 3D NMR data.Specifically, for Ubiquitin, only 10.2% of the i + 1 residues were found in the first 10 results using the default options, while 40.5% were located between the 10th and 40th results, and 49.3% were located beyond the top 40 results.Similarly, for Nsp7, we observed a slight improvement compared to the default options, with 20.3% of the i + 1 residues found in the first 10 results, 39.2% between the 10th and 40th results, and 40.5% located beyond the top 40 results.These findings suggest that SSIM may be more appropriate for identifying similar strips using 3D NMR data, as the PCA algorithm does not provide desirable results.

Conclusion
In this study, we evaluated the effectiveness of TINTO, a computer vision tool, in predicting the location of similar strips to aid in multidimensional NMR spectra assignment of large biomolecules.Our results demonstrate that TINTO is a promising tool for accurate NMR assignment without the need for peak picking.However, we observed that the accuracy of TINTO's predictions may vary depending on the protein being analyzed.Adjustment of a view focusing on the signals to search were found to improve the accuracy of predictions by reducing noise and clutter in the input Fig. 6 TINTO 3D results tested using Nsp7 and Ubiquitin NMR data.Both proteins data were tested using 2 variations.SSIM + Default settings and PCA + Default settings.The colors in the heatmap represent the rank of the i + 1 value in the list of results, with lower values indi-cating a more efficient method.The ideal rank is 1 (green), while the maximum rank is 40 (red).We represent the results for APES/I-PINE in colors white and black, white means that the residue was found and black means that the residue was not found with this method data.We also observed that Nsp7, a larger protein with experimental data, had a slightly lower accuracy rate than Ubiquitin, which had simulated data.Our study presents an innovative computer vision approach as an implementation for the first time to help in the assignment of multidimensional NMR spectra of large biomolecules.We adapted TINTO in an interactive way to the strip plot window inside POKY for 3D spectra assignment and as a GUI inside POKY for 2D spectra.We found that the structural similarity index and principal component analysis can be useful techniques for finding similar strips in challenging data sets, including those with overlapping peaks.However, the effectiveness of each method may depend on the specific application and dataset.
Despite not being an automatic assignment package, TINTO serves as a valuable support tool to accelerate the assignment process, particularly when handling challenging data.As such, we propose TINTO as a complementary tool to be used alongside automatic peak picking and assignment methods.Nevertheless, we acknowledge the pressing need for fully automating the assignment process.To address this, we have plans to integrate TINTO into I-PINE; however, further research is required to optimize these techniques for diverse data types and identify additional factors that may impact their performance.
To facilitate the process for the user, we provide two video tutorials, and we recommend the use of TINTO together with additional tools inside POKY, including Versatile Assigner (two-letter-code "va") (Manthey et al. 2022) and I-PINE for POKY (two-letter-code "ep"), to achieve efficient and successful assignment.Our study provides a promising approach for semi-automatic assignment and demonstrates the potential of TINTO for multidimensional NMR data.
In conclusion, the development of automatic and semiautomatic assignment tools has greatly improved the efficiency and accuracy of NMR resonance assignment, enabling researchers to study complex biomolecules with greater speed and accuracy.However, challenges remain in the assignment of complex biomolecules with overlapping resonances, and ongoing research is needed to develop new tools and techniques to overcome these challenges and further improve the accuracy and efficiency of NMR resonance assignment.
TINTO is pre-installed in the POKY suite, which is available at https:// poky.clas.ucden ver.edu.

Fig. 1
Fig. 1 Comparison of Peak-based Matching (PM) and Computer Vision (CV) for NMR Spectra Analysis.Examples of Strip Matching in Ideal and Overlapping Signal Conditions.The strips are a graphical representation of the spectra CBCACONH and HNCACB.The positive signals are represented in green color and the negatives in red color.a Shows an ideal case with good peak dispersion in the i and i + 1 residue strips, where both PM and CV methods work effectively.b and d illustrate scenarios where PM is not able to find the

Fig. 2
Fig. 2 TINTO graphical user interfaces (GUIs) for computer visionbased strip matching of 2D and 3D data.a TINTO 2D for the 2D TOCSY/NOESY combination.b The Search Settings window.It is opened when the "Preprocessing options" button is clicked.PCA and

Fig. 7 Fig. 8
Fig. 7 Nsp7 examples where peak picking-based algorithm (I-PINE) does not work.a CBCACONH strip for residue 3. b HNCACB strip for residue 3, we highlight the C alpha and beta of the residue 3. c CBCACONH strip for residue 4 showing the C alpha and beta of the i − 1 (residue 3).d HNCACB strip for residue 4, this strip should have 4 peaks corresponding to the C alpha and beta or the residues 4 and 3, we only see 2, in this case there is overlap of signals, which produce a failure in I-PINE when assigning the i − 1 or the i + 1 residues.e CBCACONH strip for residue 5. We show the C alpha and beta for the residue number 4. f HNCACB strip for residue 5.There are so many peaks in the same HN chemical shift, which makes really complicated to define which peaks correspond to the residue 5, in this case I-PINE also fails

Table 1
Protein NMR data used to test TINTO The simulated data set (ubiquitin) was generated using POKY's SIM-PROC module (two-letter-code "me")The table contains the Name of the proteins in study, the corresponding BMRB code, the Tool tested in each case, the Sequence Length of amino acids, the NMR Experiments used to test the tools, and the Data types of the Experiments

Table 2
Parameters and variations used to test TINTO 2D and TINTO 3D NC means that the option was not applied and Checked means that the option was applied when running TINTO