Among application-oriented universities, there are problems of incomplete teaching staff construction, emphasis of student training, lacking learning motivation, and lack of cooperation and exchange platform. Based on analyse of current situation of artificial intelligent talents cultivation, it was proposed about problems of improvement of teachers’ ability and student training, and targeted solutions were advised. It was thought that goals of AI talents cultivation were the developer teacher and experienced and applied student, whose development paths were established, which can be carried on by coordinated development of teachers and students. The cultivation system was structured by innovative undertaking platform construction of teachers and students, team selection mechanism, diversified study driving, building a teaching team of teachers and students, and capability development drived by integration of expertise and innovation.

Ribonucleic acid (RNA) foundation models (FMs) are reshaping computational biology, using self-supervised pretraining on massive transcriptomic corpora to acquire generalisable molecular representations. Developed initially for sequence annotation and expression prediction, they are increasingly applied to plant science, supporting tasks from gene function inference to regulatory element discovery and stress-response modelling in both model and orphan crop species. This review surveys RNA FMs with a focus on plant-relevant applications. We catalogue the major model families, including DNABERT-derived RNA adaptations, PlantRNA-FM, and Nucleotide Transformer variants, comparing their pretraining objectives, input modalities, and evaluation protocols. We examine how these models are deployed across plant genomics, transcriptomics, and functional annotation pipelines, highlighting the opportunities and the constraints arising from species-specific data gaps, limited experimental validation, and interpretability challenges. Across applications, the decisive constraint is no longer architecture but data: corpus breadth, condition-resolved benchmarks, and experimentally anchored validation now dominate the marginal gains available from model design. We close with the key open questions and a roadmap for integrating RNA FMs into the plant systems biology toolkit, emphasising benchmark design, cross-species transfer, and experimentally grounded validation. Throughout, every claim about what a model can do is held to three questions of evidence: what data the model was trained on, which benchmark supports the claim, and whether experimental validation in a plant system exists.

Fast and accurate composition-to-property prediction is critical for accelerating the discovery of multicomponent bio-adhesive hydrogels, yet current surrogate models are predominantly evaluated under random data splits that do not reflect the distribution shift encountered in real-world iterative screening. Moreover, existing deep learning approaches for small-sample formulation data offer little systematic guidance on architecture design and often depend on laboriously engineered interaction features. This study introduces Simplex-Inspired Multimodal Predictive Learning Encoder with X-modal Fusion (SIMPLEX), a dual-modality deep encoder purpose-built for composition-driven property estimation. Its gated multimodal fusion component autonomously learns inter-component interactions without hand-crafted quadratic terms. We establish a fully reproducible benchmark covering eight models with unified hyper-parameter tuning and statistical protocols. Evaluated on 316 internal hydrogel formulations under grouped cross-validation, SIMPLEX achieves first-tier predictive performance, competitive with the strongest classical baselines. On 25 prospectively collected formulations acquired from a Bayesian-optimized closed-loop campaign, a setting that introduces a deliberate distribution shift toward high-performance compositions, SIMPLEX delivers the highest coefficient of determination ( R 2 = 0.6712 ) for adhesion-strength prediction, meeting the demand for precise quantitative estimation in iterative candidate screening. Feature dissection reveals that the native compositional representation already captures the dominant composition–property relationships, and component-wise ablation pinpoints gated multimodal fusion as the principal performance driver. This study delivers a capable deep encoder together with a rigorous, transparent evaluation paradigm for formulation-driven machine learning, providing immediately reusable baselines and design principles for the community.

Plant growth-promoting rhizobacteria (PGPR) can support crop growth and protection, but their performance varies across strains, hosts, soils, and management conditions. Colonization measurements can help explain this variation, although persistence, microbial function, and plant benefit are distinct outcomes. This critical review examines how genome-informed models could prioritize strains for defined colonization tasks. We connect the ecology of dispersal, attachment, competition, and persistence with phenotype measurements, sequence representations, and validation design. Protein language models, DNA-based encoders, pan-genome features, and multi-task learning provide candidate tools, supported mainly by applications to related prediction problems. Their value for strain-level persistence must be assessed against phenotype-based and simpler genomic baselines under matched evaluation conditions. We organize the evidence around three application directions, examine assumptions about laboratory-to-field transfer, functional annotation, and trait independence, and propose a metadata schema and benchmark framework. Independent strain and soil holdouts, endpoint-specific measurements, and prospective field studies are central to assessing added predictive value. The resulting framework is a proposal for testing when genomic information and model complexity improve strain prioritization, with scope determined by the available labels and validation setting.

Structural topology optimization of large-scale engineering systems requires repeated compliance evaluations, making conventional finite element solvers computationally prohibitive in iterative design workflows. Physics-informed neural networks (PINNs) offer a mesh-free, differentiable surrogate alternative, but standard PINNs yield unstable sensitivity fields due to error amplification in higher-order automatic differentiation, undermining optimization reliability. This paper introduces a gradient-enhanced PINN surrogate (gSurrogate) that incorporates a residual-gradient penalty into the total potential energy loss, directly regularizing displacement field derivatives and stabilizing compliance sensitivities for practical structural design. The surrogate is validated on cantilever, simply supported, and fixed–fixed beam configurations and embedded within a SIMP-based topology optimization pipeline driven by the Method of Moving Asymptotes. Over 200 optimization iterations, the proposed surrogate reduces unfiltered sensitivity variance by 94.35\% and filtered variance by 81.74\% relative to a standard PINN, with variance standard deviation reduced by 97.16\%. Three-dimensional topology optimization across mesh resolutions of 120×20×20 and 60×60×20 demonstrates that the gSurrogate achieves displacement errors of 1.81\% and 1.63\% with respect to FEM (Abaqus), compared to 3.75\% and 3.45\% for a standard PINN, a consistent twofold accuracy improvement. These results establish the proposed surrogate as a computationally efficient and numerically reliable alternative to finite element solvers for compliance-driven topology optimization of structural systems.