The AI-Driven Blueprint: Decoding Intervertebral Disc Repair Mechanisms for Intelligent Biomaterial Design

Yuanzhen Shi¹Yifan Wang^2*

• ¹The Second Hospital of Shandong University, Jinan, China
• ²People's Liberation Army General Hospital, Beijing, China

The intervertebral disc (IVD), serving as a critical load-bearing structure connecting vertebral bodies, maintains spinal mechanical stability and function through the integrated health of its nucleus pulposus, annulus fibrosus, and cartilaginous endplates (Knezevic et al., 2021). However, intervertebral disc degeneration (IVDD) has emerged as a global health challenge, with its prevalence increasing markedly with age. As the leading cause of chronic low back pain, it imposes a substantial healthcare burden on society. Current clinical strategies for managing IVDD primarily focus on symptomatic relief-including physical therapy, pain management, and spinal fusion surgeries aimed at mechanical stabilization-none of which fundamentally reverse the degenerative process or achieve functional tissue regeneration (Copeland, 2007;Binch et al., 2021).With the expanding integration of artificial intelligence (AI) into medical science, particularly its growing capacity to uncover profound patterns within complex biomedical data, a new paradigm has emerged to address this challenge (Zhang et al., 2025). The application of AI in elucidating the mechanisms of IVDD and designing corresponding repair materials offers novel perspectives for regenerative rehabilitation of degenerative discs. Artificial intelligence and machine learning are rapidly transforming intervertebral-disc research by integrating multi-omics data, single-cell profiles, and in-silico drug screening to decode the molecular logic of disc degeneration and repair. These algorithms outperform conventional statistics in ranking non-coding RNAs, programmed-cell-death regulators, or immune-matrix crosstalk genes, providing clinicians with quantitative biomarkers and druggable targets that can be validated in vitro and in rodent models (Liao et al., 2025).The application of AI now extends to automating the diagnosis and grading of disc degeneration from lumbar MR images, where deep learning models demonstrate high precision in identifying pathological changes. Furthermore, single-cell RNA sequencing (scRNA-seq) has become a pivotal tool, revealing transcriptional shifts in disc resident and infiltrating cell populations following injury, and uncovering novel cellular targets for repair strategies. For instance, specific mesenchymal stem cell (MSC) populations were identified whose differentiation shifts with injury, offering potential for regenerative therapies (Lin et al., 2024;Stirnimann et al., 2025).Li et al introduced an RNA-seq-driven competing-endogenous-RNA strategy that couples differential lncRNAs, miRNAs and mRNAs in degenerated versus traumatic discs; they experimentally confirmed the XIST-miR-4775-PLA2G7 and XIST-miR-424-AMOT/TGFBR3 axes as pro-inflammatory circuits that disrupt extracellularmatrix homeostasis (Li et al., 2021). Zhang et al presented a bioinformatics-plusmachine-learning pipeline that merges four GEO microarray sets, performs WGCNA-LASSO screening and constructs a protein-protein interaction network, identifying IL1R1 and TCF7L2 as central transcriptional hubs whose elevated expression distinguishes advanced Pfirrmann-grade discs with an AUC ≈ 0.7 (Zhang et al., 2024).Lv et al delineated a comprehensive programmed-cell-death atlas of IVDD by integrating bulk and single-cell transcriptomes; machine-learning models selected PDCD6 and UBE2K as apoptosis-specific drivers, and in vivo administration of the repositioned drug Glibenclamide attenuated caspase-3 activity, preserved disc height and validated the predictive value of their ridge-regression score (Lv et al., 2025). AI-guided single-cell multi-omics is advancing the rational design of biomaterials for intervertebral disc repair. By mapping cellular heterogeneity, cell death pathways, and fibrotic signatures at single-cell resolution, machine learning models enable the identification of therapeutic targets that are inaccessible via bulk profiling. These insights inform the engineering of mitochondria-targeting carriers, siRNA nanomotors, and immunomodulatory scaffolds-each mathematically optimized to maximize onsite efficacy while minimizing off-target effects. Convolutional neural networks cluster cell populations into distinct transcriptional states, whereas trajectory inference and random forest models predict lineage plasticity and gene significance. Meanwhile, reinforcement learning simulates the spatiotemporal behavior of biomaterials within a reconstructed disc microenvironment, iteratively refining their physical and biochemical properties to achieve an optimal therapeutic profile. This AI-driven approach transforms traditional "make-and-test" cycles into an in-silico "predictdesign-validate" workflow, accelerating development, reducing animal use, and enabling patient-specific implants (Gao et al., 2022;Hu et al., 2023).Tu et al presented an atlas of human nucleus pulposus at single-cell resolution that resolves six NPC sub-states, fibro-progenitor trajectories and CD90+ progenitors with tri-lineage potency; bioinformatic deconvolution of immune infiltration revealed G-MDSC-mediated immunosuppression, inspiring a progenitor-enriched hydrogel that arrests fibrosis in rat IVDD (Tu et al., 2022). Zhou et al introduced Motor@TA-siRNA, an H2O2-propelled nanomotor whose trajectory, siRNA load and tannic-acid shield were optimized by single-cell evidence of STING-driven pyroptosis; the carrier selfenriches within fibrotic NP, silences STING and simultaneously scavenges ROS, doubling disc height retention relative to passive vectors (Zhou et al., 2025) (Figure 1).Yang et al described a deep-learning-assisted mitochondrial therapy in which scRNAseq pinpointed OXPHOS-deficient fibrochondrocytes as the prime driver of fibrosis; exogenous mitochondria engineered with a mitochondria-targeting macromolecule PSP were shown to restore respiration, block mtDNA leakage and disrupt the SPARC-STING axis, achieving near-native NP architecture in a rat puncture model (Yang et al., 2025). In our view, the full integration of AI into future intervertebral disc repair strategies is not only highly valuable but also an indispensable component. Conventional approaches struggle to address the complex etiological factors and significant interindividual variability of IVDD. In contrast, AI provides a robust, data-driven framework for developing personalized therapeutic regimens grounded in solid mechanistic foundations. We should strive to build multimodal AI systems that integrate biomechanical, imaging, and clinical data to simulate the processes of disc degeneration and repair within a holistic digital twin framework. On the basis of interdisciplinary collaboration, we will advance these technological breakthroughs from theoretical concepts to clinical practice.