With advancements in sequencing and mass spectrometry technologies, multi-omics data can now be easily acquired for understanding complex biological systems. Nevertheless, substantial challenges remain in determining the association between gene-metabolite pairs due to the complexity of cellular networks. Here, we introduce Compounds and Transcripts Bridge (abbreviated as CAT Bridge, freely available at http://catbridge.work), a user-friendly platform for longitudinal multi-omics analysis to efficiently identify transcripts associated with metabolites using time-series omics data. To evaluate the association of gene-metabolite pairs,

causality estimation and correlation coefficient calculation for multi-omics analysis.

interpreting the association results. We applied CAT Bridge to self-generated (chili pepper) and public (human) time-series transcriptome and metabolome datasets. CAT Bridge successfully identified genes involved in the biosynthesis of capsaicin in Capsicum chinense. Furthermore, case study results showed that the convergent cross mapping (CCM) method outperforms traditional approaches in longitudinal multi-omics analyses. CAT Bridge simplifies access to various established methods for longitudinal multi-omics analysis, and enables researchers to swiftly identify associated gene-metabolite pairs for further validation.

metabolites, correlation is often used to imply association, such as Spearman Correlation Coefficient (Spearman) and Pearson Correlation Coefficient (Pearson). However, such correlation-based methods have substantial limitations (7,8) because they tend to overlook the non-linearity and lag issues of gene expression leading to metabolite changes in complex biological systems. Therefore, besides considering Spearman and Pearson, we have integrated into CAT Bridge various distinct statistical methods, including Convergent Cross Mapping (CCM) and Granger Causality (Granger) to identify causality between gene-metabolite pairs, as well as Canonical Correlation Analysis (CCA), Dynamic-Time-Warping (DTW), CrossCorrelation Function (CCF) for calculating correlation coefficient (The implementation methods are provided in the Supplementary text). These algorithms were based on different assumptions, so that some of them allow compatibility with time series data and complex systems. Among them, correlation-based strategies have been widely applied in genomics and multi-omics analysis (11-14). The CCM and Granger, which estimate causality fromtime series data, are already used in some areas of biology such as ecology and neurobiology but leave a gap in the omics analysis (15-18). The integration of diverse computational methods provides users with the flexibility to select the approach that best suits their needs. Furthermore, our benchmarking results (as detailed in the Results section) suggest that in longitudinal multiomics studies, causal relationships may provide a more accurate depiction of the association between genes and metabolites.

Fold change (FC) is another measurement frequently used in omics analyses to identify differentially expressed genes (19). CAT Bridge pinpoints the peak time of the ta rget metabolite and calculates each gene's log2 normalized FC of this peak time point and the subsequent decline time point (that is, the next sampling time point after the peak time point) using DESeq2 (20). Then, causality/correlation and FC are combined into a vector to represent the gene-metabolite pair. After the min-max normalization of values (The details are provided in the Supplementary text), the magnitude of this vector is calculated as the CAT score (Figure

Optionally, if users provide a gene function annotations file, typically derived from homology annotations using tools like InterProScan (21) or eggNOG-mapper (22) for non-model organisms), the CAT score will be adjusted with an additional value. This value is determined by a scoring rule based on the gene's description. By default, genes identified as enzymes receive a score of 0.2, while those with unknown functions get a score of 0.1. Users can customize this scoring rule based on their specific requirements, depending on the presence of target annotations and their importance.

non-model organisms, we collected transcriptome sequencing and metabolic profiling data from chili peppers (Capsicum chinese L.), focusing on one of its trademark natural products, capsaicin. Pepper seedlings were grown in a greenhouse of Peking University Institute of Advanced Agricultural Sciences with a controlled environment of 25°C temperature, a lightdark cycle of 16 hours light and 8 hours dark, and 70% relative humidity. The fruits of peppers were sampled at seven distinct time points, starting from the day of flowering, i.e. 0 day postanthesis (DPA) during which flowers were collected. Following this, fruits were harvested on days 7, 16, 30, 50, 55, and 60 DPA. The samples were ground and freeze-dried in liquid nitrogen, followed by extraction using 1.0 mL of 70% aqueous methanol for every 50 mg of the sample, and an ultrasonic process was employed for 30 minutes. The preparation of standards was conducted as follows: a mixed standard in the range of 20-50¿g/mL was prepared using methanol of MS grade. For the amino acid standard solution, a 1mg/mL stock solution was prepared in water and subsequently diluted with 50% methanol to achieve a concentration of 50¿g/mL. Three biological replicates were used in later transcriptome and metabolome analysis.

chromatography coupled with mass spectrometry (LC-MS). The samples were filtered through a 0.22 µm membrane and transferred into the lining tube of a sampling vial. Subsequent centrifugation was carried out at 12000 rcf and 4# for 10 minutes. The processed samplse were then analyzed using Thermo Scientific Orbitrap Exploris# 240 (Thermo FisherScientific, USA). Chromatographic separation was achieved on a T3 C18 (1.7 µm, 2.1 mm × 150 mm column, USA) maintained at 40#. The mobile phase consisted of A: 1% formic caid in water and B: 1% formic acid in acetonitrile, with a flow rate of 300 µL/min. A 3 µL sample was injected at an autosampler temperature of 10#. The elution gradient was set as follows: 0-2.5 95% B; 23-23.1 min, 95-3% B; 23.1-28 min, 3% B. MS was performed using both positive and negative ion scans, with a precursor ion scan mode. The auxiliary gas heater temperature was set at 350#, and the ion transfer tube temperature was also maintained at 350#. The sheath gas flow rate and auxiliary gas flow rate were set to 35 arb and 15 arb, respectively. The voltages were set to 3.5 KV for the positive spectrum and 3.2 KV for the negative spectrum.

evaluated using RNA Nano 6000 Assay Kit of the Bioanalyzer 2100 system (Agilent

value >7 were used in downstream sequencing library construction and sequencing. The library construction was conducted using Illumina True-seq transcriptome kit (Illumina, USA) following standard protocols. Transcriptome sequencing was carried out by Novogene Co., Ltd.

processing, fastp (24) was employed to conduct quality control and clean the data.

performance of CAT Bridge. The data was sourced from a published precision medicine study (28) that conducted a deep longitudinal multi-omics analysis of 105 individuals over four years. We downloaded the transcriptomic and metabolomic data from the dataset, and generated functional annotations for each gene using Entrez Programming Utilities. Then we chose the participant with the most extensive time points to provide the most time points. Furthermore, as the intervals of the sampling times were not uniform, only time points where both transcriptomics and metabolomics were sampled were retained, and only the earliest time point for the month was preserved if more than one time point was sampled in one month. Then, we calculated the slopes for all metabolites using linear regression to select target metabolite for the following analysis.

all users without any login requirements. This is particularly beneficial to those who are less familiar with programming language. Secondly, a standalone application is available for users handling large data files, as it lifts the constraints on file sizes. Finally, a Python library is available for bioinformaticians with complete features and customizable workflows (The implementation of the software is provided in the Supplementary text).

dataset collected from chili peppers and a publicly available human dataset in two case studies.

tend to outperform those solely based on correlation. As such, we advocate for the adoption of causality rather than correlation in longitudinal multi-omics analysis such as co-mining the transcriptomics and metabolomics data. The capsaicin dataset generated in this study has been made open source for user exploration.

method for hypothetical ranking, BC332_05016 encoding an Acyl-transferase was ranked first, regardless of whether an annotation file was provided. The Result suggests that this gene was more likely to be the synthetic gene associated with capsaicin in C. chinense (Figure 2B). The complete heuristic ranking results are provided in Supplementary Table S1. BLAST search revealed that BC332_05016 was homologous of PUN1 (sequence identity: 100%), a.k.a AT3 (Acyl-transferase 3) or CS (capsaicin synthesis) gene (30). Moreover, when common thresholds were applied for screening, only CCM passed the criteria. The causality modeled based on CCM was 0.55, implying a strong association between BC332_05016 and capsaicin. By contrast, the conventional Pearson correlation method produced a result of 0.08, which would fall below the commonly used threshold and potentially lead to an overlook of this gene-metabolite pair (Figure 2C). The AI agent also accurately found BC332_05016 among the top 100 genes based on functional annotation (Figure 2D). Furthermore, CAT Bridge visualization result showed that capsaicinoids such as nonivamide, dihydrocapsaicin, and homocapsaicin have high similarity to capsaicin (Figure 2E) and may play a significant role in response to the variable (Figure 2F). The rest of the visualization results are provided in the Supplementary text. These results show that the CAT

fruit development of chili peppers in case study 1. (B) Heuristic ranking produced using the CCM-based method. (C) Comparative values across different methods. For unnormalized values: red indicates a strong association; original denotes medium association; blue suggests values that are below the commonly used threshold, show no association, or are negatively associated (depending on the method); light blue means this method does not adhere to a common threshold. For normalized values: red signifies values that are high after min-max normalization; blue represents low normalized values. (D) Interpretation results derived from the AI agent (E) The correlation network of capsaicin. (F) The significance of metabolites.

can methylate guanidinoacetate to creatine, and creatine spontaneously converts to creatinine subsequently (31). And the consistent reduction in urinary creatinine excretion is an unspecific indicator of GAMT deficiency (32). These findings indicate that GAMT is a key gene involved in creatinine regulation. The result from the CAT Bridge revealed that 61 genes, including

and FC (complete heuristic ranking results are provided in Supplementary Table S2). However, computed values of GAMT-creatinine from other methods indicate a lack of association (Figure 3B). The AI agent also successfully identified GAMT as a candidate gene among the top 100 genes based on domain knowledge (Figure 3C). Interestingly, even if the values for

values after min-max normalization are higher. The complexities inherent in human regulatory mechanisms and the stability of molecular levels during the sampling timeframe might account for this observation. In addition, creatinine is also shown as an important determinant of variance in the metabolome (Figure 3D).

These results demonstrate the CAT Bridge's potential to extract meaningful insights from multi-omics data across diverse species. By integrating time-series analysis methods, particularly CCM, it offers superior performance in longitudinal omics compared to common methods. (A) Concentration of creatinine across sampling time points. The color of the dots indicates the season: green for spring, red for summer, yellow for fall, and blue for winter. (B) Comparative values across different methods. For color interpretation, refer to the annotation in Figure 2. (C) Interpretation results derived from the AI agent. (D) The significance of the metabolites.

CorDiffViz integrate either Pearson or Spearman correlations, or both, to aid in the discovery of feature relationships. What sets CAT Bridge apart is its assembly of various algorithms that handle time-series data and causality, and incorporate an AI agent to inspire users. Notably, the performance of CCM has been found from two previous case studies to be potentially more suitable for longitudinal multi-omics analysis compared to traditional methods. × indicating feature assessments: '7' denotes presence, ' ' signifies absence.

analytics

association

platform

7 7 7 7 7 7 7 7 × × × × × 7 7 × × × × 7 7 7 × × × 7 7 7 × ×

overlooked in such studies is the unique nature of the longitudinal experimental design. Longitudinal omics analysis is particularly important in research on the developmental cycle of plants and investigations related to chronic diseases and aging (36-39). However, ma ny studies tend to use generic methodologies for analysis (9,10). This may inadvertently miss key discoveries. CAT Bridge provides a platform specifically for longitudinal multi-omics analysis, by drawing insights from disciplines where time series data is more prevalent, and validating them

with data. Through the gene-metabolite causality/correlation modeling method, combined with visualization tools and AI assistance, researchers can more quickly identify putative genes for experimental validation.

In two case studies, CCM showed its superiority compared to other methods, which is probably due to its modeling capability of dynamical systems. Thus, it can better capture the complex non-linear interactions within biological systems(15). We advocate for modeling cause-andeffect relationships in longitudinal omics analyses, instead of more widely used Pearson or

Aside from computational methods, the reliability of analytical results is also influenced by experimental design and data acquisition methods. Increasing the number of sampling time points and setting a reasonable interval between them can enhance the power of association and credibility of the results. On the data acquisition front, it is recommended to annotate the transcriptome with an updated, high-quality reference genome, and employ advanced metabolomics techniques such as chemical isotope labeling liquid (CIL) LC-MS (40) to ensure a high coverage and more accurate relative quantification of metabolome.

and Natural Science Foundation for Distinguished Young Scholars (ZR2023JQ010) of

18, 83.

Hasin, Y., Seldin, M. and Lusis, A. (2017) Multi -omics approaches to disease. Genome Biology, Subramanian, I., Verma, S., Kumar, S., Jere, A. and Anamika, K. (2020) Multi-omics Data Integration, Interpretation, and Its Application. Bioinform Biol Insights, 14, 1177932219899051. Zhou, G., E wald, J. and Xia, J. (2021 ) OmicsAnal yst: a comprehensive web-based platform for visual analytics of multi-omics data. Nucleic Acids Research, 49, W476-W482. Krassowski, M., Das, V., Sahu, S.K. and Misra, B.B. (2020) State of the Field in Multi-Omics Research: From Computational Needs to Data Mining and Sharing. Front Genet, 11, 610798. Athieniti, E. and Spyrou, G.M. (2023) A guide to multi-omics data collection and integration for translational medicine. Computational and Structural Biotechnology Journal, 21, 134-149. Cavill, R., Jennen, D., Kleinjans, J. and Briedé, J.J. (2015) Transcriptomic and metabolomic data integration. Briefings in Bioinformatics, 17, 891-901.

Chong, J. and Xia, J. (2017) Computational Appro aches for Integrative Analysis of the Metabolome and Microbiome. Metabolites, 7, 62.

Li, Y., Chen, Y., Zhou, L., You, S., Deng, H., C hen, Y., Alseekh, S., Yuan, Y., Fu, R. and Zhang, Z. (2020) MicroTom metabolic network: rewiring tomato metabolic regulatory network throughout the growth cycle. Molecular plant, 13, 1203-1218. 10.

Yang, C., Shen, S., Zhou, S., Li, Y., Mao, Y., Zhou, J., Shi, Y., An, L., Zhou, Q. and Peng, W. (2022) Rice metabolic regulatory network spanning the entire life cycle. Molecular Plant, 15, 258-275.

Rockwood, A.L., Crockett, D.K., Oliphant, J.R. and Elenitoba-Johnson, K.S. (2005) Sequence alignment by cross-correlation. J Biomol Tech, 16, 453-458.

Skutkova, H., Vitek, M., Babula, P., Kizek, R. and Provaznik, I. (2013) Classification of genomic signals using dynamic time warping. BMC Bioinformatics, 14, S1.

Seoane, J.A., Campbell, C., Day, I.N., Casas, J .P. and Gaunt, T.R. (2014) Canonical correlation

analysis for gene-based pleiotropy discovery. PLoS Comput Biol, 10, e1003876. 14.

Jiang, M.Z., Aguet, F., Ardlie, K., Chen, J., C ornell, E., Cruz, D., Durda, P., Gabriel, S.B., series. eLife, 11, e72518. irregularly sampled time series with application to nitrogen signalling in Arabidopsis. Bioinformatics, 37, 2450-2460.

Ye, H., Deyle, E.R., Gilarranz, L.J. and Sugiha ra, G. (2015) Distinguishing time-delayed causal interactions using convergent cross mapping. Scientific Reports, 5, 14750.

Stokes, P.A. and Purdon, P.L. (2017) A study of problems encountered in Granger causality analysis from a neuroscience perspective. Proc Natl Acad Sci U S A, 114, E7063-e7072. Arora, S., Pattwell, S.S., Holland, E.C. and Bo louri, H. (2020) Variability in estimated gene expression among commonly used RNA-seq pipelines. Scientific Reports, 10, 2734. Love, M.I., Huber, W. and Anders, S. (2014) Mod erated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biology, 15, 550.

Jones, P., Binns, D., Chang, H.-Y., Fraser, M., Li, W., McAnulla, C., McWilliam, H., Maslen, J., Mitchell, A., Nuka, G. et al. (2014) InterProSc an 5: genome-scale protein function classification. Bioinformatics, 30, 1236-1240. eggNOG-mapper v2: Functional Annotation, Orthology Assignments, and Domain Prediction at the Metagenomic Scale. Molecular Biology and Evolution, 38, 5825-5829.

Kumar, L. and M, E.F. (2007) Mfuzz: a software package for soft clustering of microarray data. Bioinformation, 2, 5-7.

Chen, S., Zhou, Y., Chen, Y. and Gu, J. (2018) fastp: an ultra-fast all-in-one FASTQ preprocessor. Bioinformatics, 34, i884-i890.

Kim, S., Park, J., Yeom, S.I., Kim, Y.M., Seo, E., Kim, K.T., Kim, M.S., Lee, J.M., Cheong, K., Shin, H.S. et al. (2017) New reference genome sequences of hot pepper reveal the massive evolution of plant disease-resistance genes by retroduplication. Genome Biol, 18, 210. 26. 27. (2015) StringTie enables improved reconstruction of a transcriptome from RNA-seq reads.

Nature Biotechnology, 33, 290-295.

Sailani, M.R., Metwally, A.A., Zhou, W., Rose, S.M.S.-F., Ahadi, S., Contrepois, K., Mishra, T., Zhang, M.J., KidziEski, A., Chu, T.J. et al. (2020) Deep longitudinal multiomics profiling reveals two biological seasonal patterns in California. Nature Communications, 11, 4933. Fattori, V., Hohmann, M.S., Rossaneis, A.C., Pi nho-Ribeiro, F.A. and Verri, W.A. (2016) Capsaicin: Current Understanding of Its Mechanisms and Therapy of Pain and Other PreClinical and Clinical Uses. Molecules, 21.

Kim, S., Park, M., Yeom, S.-I., Kim, Y.-M., Lee, J.M., Lee, H.-A., Seo, E., Choi, J., Cheong, K., Kim, K.-T. et al. (2014) Genome sequence of the hot pepper provides insights into the evolution of pungency in Capsicum species. Nature Genetics, 46, 270-278. 31.

da Silva, R.P., Nissim, I., Brosnan, M.E. and B rosnan, J.T. (2009) Creatine synthesis: hepatic metabolism of guanidinoacetate and creatine in the rat in vitro and in vivo. Am J Physiol Endocrinol Metab, 296, E256-261. 32.

Stockler-Ipsiroglu, S., van Karnebeek, C., Longo, N., Korenke, G.C., Mercimek-Mahmutoglu, S., Marquart, I., Barshop, B., Grolik, C., Schlune, A., Angle, B. et al. (2014) Guanidinoacetate methyltransferase (GAMT) deficiency: Outcomes in 48 individuals and recommendations for diagnosis, treatment and monitoring. Molecular Genetics and Metabolism, 111, 16-25. 33.

Kuo, T.C., Tian, T.F. and Tseng, Y.J. (2013) 3Omics: a web-based systems biology tool for analysis, integration and visualization of human transcriptomic, proteomic and metabolomic data. BMC Syst Biol, 7, 64. 34.

Siddiqui, J.K., Baskin, E., Liu, M., Cantemir-S tone, C.Z., Zhang, B., Bonneville, R., McElroy, J.P., Coombes, K.R. and Mathé, E.A. (2018) IntLIM: integration using linear models of metabolomics and gene expression data. BMC Bioinformatics, 19, 81. 35.

Yu, S., Drton, M., Promislo w, D.E.L. and Shojai e, A. (2021 ) CorDiffViz: an R package for visualizing multi-omics differential correlation networks. BMC Bioinformatics, 22, 486. back pain: protocol of a retrospective longitudinal study. BMJ Open, 6, e012070. Group-Submetabolome Analysis: Group Classification and Four-Channel Chemical Isotope Labeling LC-MS. Analytical Chemistry, 91, 12108-121 15.