Jiaqi Wei

Xiang Zhang, Jiaqi Wei, Yuejin Yang et al.

Chain-of-Thought (CoT) prompting has significantly advanced task-solving capabilities in natural language processing with large language models. Unlike standard prompting, CoT encourages the model to generate intermediate reasoning steps, non-answer tokens, that help guide the model toward more accurate final outputs. These intermediate steps enable more complex reasoning processes such as error correction, memory management, future planning, and self-reflection. However, applying CoT to non-natural language domains, such as protein and RNA language models, is not yet possible, primarily due to the limited expressiveness of their token spaces (e.g., amino acid tokens). In this work, we propose and define the concept of language expressiveness: the ability of a given language, using its tokens and grammar, to encode information. We show that the limited expressiveness of protein language severely restricts the applicability of CoT-style reasoning. To overcome this, we introduce reflection pretraining, for the first time in a biological sequence model, which enables the model to engage in intermediate reasoning through the generation of auxiliary "thinking tokens" beyond simple answer tokens. Theoretically, we demonstrate that our augmented token set significantly enhances biological language expressiveness, thereby improving the overall reasoning capacity of the model. Experimentally, our pretraining approach teaches protein models to self-correct and leads to substantial performance gains compared to standard pretraining.

Wanghan Xu, Yuhao Zhou, Yifan Zhou et al.

Despite advances in scientific AI, a coherent framework for Scientific General Intelligence (SGI)-the ability to autonomously conceive, investigate, and reason across scientific domains-remains lacking. We present an operational SGI definition grounded in the Practical Inquiry Model (PIM: Deliberation, Conception, Action, Perception) and operationalize it via four scientist-aligned tasks: deep research, idea generation, dry/wet experiments, and experimental reasoning. SGI-Bench comprises over 1,000 expert-curated, cross-disciplinary samples inspired by Science's 125 Big Questions, enabling systematic evaluation of state-of-the-art LLMs. Results reveal gaps: low exact match (10--20%) in deep research despite step-level alignment; ideas lacking feasibility and detail; high code executability but low execution result accuracy in dry experiments; low sequence fidelity in wet protocols; and persistent multimodal comparative-reasoning challenges. We further introduce Test-Time Reinforcement Learning (TTRL), which optimizes retrieval-augmented novelty rewards at inference, enhancing hypothesis novelty without reference answer. Together, our PIM-grounded definition, workflow-centric benchmark, and empirical insights establish a foundation for AI systems that genuinely participate in scientific discovery.

Mengru Wu, Jiawei Li, Jiaqi Wei et al.

With the increasing computational demands of deep neural network (DNN) inference on resource-constrained devices, DNN partitioning-based device-edge collaborative inference has emerged as a promising paradigm. However, the transmission of intermediate feature data is vulnerable to malicious jamming, which significantly degrades the overall inference performance. To counter this threat, this letter focuses on an anti-jamming collaborative inference system in the presence of a malicious jammer. In this system, a DNN model is partitioned into two distinct segments, which are executed by wireless devices and edge servers, respectively. We first analyze the effects of jamming and DNN partitioning on inference accuracy via data regression. Based on this, our objective is to maximize the system's revenue of delay and accuracy (RDA) under inference accuracy and computing resource constraints by jointly optimizing computation resource allocation, devices' transmit power, and DNN partitioning. To address the mixed-integer nonlinear programming problem, we propose an efficient alternating optimization-based algorithm, which decomposes the problem into three subproblems that are solved via Karush-Kuhn-Tucker conditions, convex optimization methods, and a quantum genetic algorithm, respectively. Extensive simulations demonstrate that our proposed scheme outperforms baselines in terms of RDA.

Zhihan Yang, Jiaqi Wei, Xiang Zhang et al.

Tabular data high-stakes critical decision-making in domains such as finance, healthcare, and scientific discovery. Yet, learning effectively from tabular data in few-shot settings, where labeled examples are scarce, remains a fundamental challenge. Traditional tree-based methods often falter in these regimes due to their reliance on statistical purity metrics, which become unstable and prone to overfitting with limited supervision. At the same time, direct applications of large language models (LLMs) often overlook its inherent structure, leading to suboptimal performance. To overcome these limitations, we propose FORESTLLM, a novel framework that unifies the structural inductive biases of decision forests with the semantic reasoning capabilities of LLMs. Crucially, FORESTLLM leverages the LLM only during training, treating it as an offline model designer that encodes rich, contextual knowledge into a lightweight, interpretable forest model, eliminating the need for LLM inference at test time. Our method is two-fold. First, we introduce a semantic splitting criterion in which the LLM evaluates candidate partitions based on their coherence over both labeled and unlabeled data, enabling the induction of more robust and generalizable tree structures under few-shot supervision. Second, we propose a one-time in-context inference mechanism for leaf node stabilization, where the LLM distills the decision path and its supporting examples into a concise, deterministic prediction, replacing noisy empirical estimates with semantically informed outputs. Across a diverse suite of few-shot classification and regression benchmarks, FORESTLLM achieves state-of-the-art performance.

Xiang Zhang, Jiaqi Wei, Zijie Qiu et al.

Autoregressive (AR) models, common in sequence generation, are limited in many biological tasks such as de novo peptide sequencing and protein modeling by their unidirectional nature, failing to capture crucial global bidirectional token dependencies. Non-Autoregressive (NAR) models offer holistic, bidirectional representations but face challenges with generative coherence and scalability. To transcend this, we propose a hybrid framework enhancing AR generation by dynamically integrating rich contextual information from non-autoregressive mechanisms. Our approach couples a shared input encoder with two decoders: a non-autoregressive one learning latent bidirectional biological features, and an AR decoder synthesizing the biological sequence by leveraging these bidirectional features. A novel cross-decoder attention module enables the AR decoder to iteratively query and integrate these bidirectional features, enriching its predictions. This synergy is cultivated via a tailored training strategy with importance annealing for balanced objectives and cross-decoder gradient blocking for stable, focused learning. Evaluations on a demanding nine-species benchmark of de novo peptide sequencing show that our model substantially surpasses AR and NAR baselines. It uniquely harmonizes AR stability with NAR contextual awareness, delivering robust, superior performance on diverse downstream data. This research advances biological sequence modeling techniques and contributes a novel architectural paradigm for augmenting AR models with enhanced bidirectional understanding for complex sequence generation. Code is available at https://github.com/BEAM-Labs/denovo.

Chengdong Xu, Kaiqiang Ke, Ziheng Liu et al.

Large language model (LLM)-based multi-agent systems have shown strong potential on complex tasks through agent specialization, tool use, and collaborative reasoning. However, most automated multi-agent system design methods still follow a one-shot paradigm: a workflow is optimized or selected before execution and then reused unchanged throughout the task. This static coordination strategy is ill-suited for long-horizon tasks whose subgoals, intermediate evidence, and information needs evolve over multiple execution stages. We propose EvoMAS, a framework for execution-time multi-agent workflow construction. EvoMAS formulates workflow construction as a meta-level sequential decision problem along a single task trajectory. At each stage, it constructs an explicit task state through a Planner-Evaluator-Updater pipeline and uses a learned Workflow Adapter to instantiate a stage-specific layered workflow from a fixed pool of candidate agents. The adapter is trained with policy gradients using sparse, verifiable terminal task success as the main supervision signal, while evaluator-based process reward is analyzed separately under very-hard sparse-reward settings. Experiments on GAIA, HLE, and DeepResearcher show that EvoMAS outperforms single-agent baselines and recent automated multi-agent workflow design methods. Our analyses further show that explicit task-state construction and learned workflow adaptation provide complementary benefits. Additional results indicate that process reward is most useful when terminal success is extremely sparse, and qualitative case studies illustrate that EvoMAS adapts agent coordination as the task state evolves.

Jiaqi Wei, Xuehang Guo, Pengfei Yu et al.

In single-stream autoregressive interfaces, the same tokens both update the model state and constitute an irreversible public commitment. This coupling creates a \emph{silence tax}: additional deliberation postpones the first \emph{task-relevant} content, while naive early streaming risks premature commitments that bias subsequent generations. We introduce \textbf{\emph{Side-by-Side (SxS)}} Interleaved Reasoning, which makes \emph{disclosure timing} a controllable decision within standard autoregressive generation. SxS interleaves partial disclosures with continued private reasoning in the same context, but releases content only when it is \emph{supported} by the reasoning so far. To learn such pacing without incentivizing filler, we construct entailment-aligned interleaved trajectories by matching answer prefixes to supporting reasoning prefixes, then train with SFT to acquire the dual-action semantics and RL to recover reasoning performance under the new format. Across two Qwen3 architectures/scales (MoE \textbf{Qwen3-30B-A3B}, dense \textbf{Qwen3-4B}) and both in-domain (AIME25) and out-of-domain (GPQA-Diamond) benchmarks, SxS improves accuracy--\emph{content-latency} Pareto trade-offs under token-level proxies (e.g., inter-update waiting).

Wenxuan Huang, Mingyu Tsoi, Yanhao Huang et al.

Single-cell perturbation studies face dual heterogeneity bottlenecks: (i) semantic heterogeneity--identical biological concepts encoded under incompatible metadata schemas across datasets; and (ii) statistical heterogeneity--distribution shifts from biological variation demanding dataset-specific inductive biases. We propose HarmonyCell, an end-to-end agent framework resolving each challenge through a dedicated mechanism: an LLM-driven Semantic Unifier autonomously maps disparate metadata into a canonical interface without manual intervention; and an adaptive Monte Carlo Tree Search engine operates over a hierarchical action space to synthesize architectures with optimal statistical inductive biases for distribution shifts. Evaluated across diverse perturbation tasks under both semantic and distribution shifts, HarmonyCell achieves a 95% valid execution rate on heterogeneous input datasets (versus 0% for general agents) while matching or even exceeding expert-designed baselines in rigorous out-of-distribution evaluations. This dual-track orchestration enables scalable automatic virtual cell modeling without dataset-specific engineering.

Xiang Zhang, Juntai Cao, Jiaqi Wei et al.

Despite the remarkable successes of large language models (LLMs), the underlying Transformer architecture has inherent limitations in handling complex reasoning tasks. Chain-of-thought (CoT) prompting has emerged as a practical workaround, but most CoT-based methods rely on a single, generic prompt such as "think step by step", with no task-specific adaptation. These approaches expect the model to discover an effective reasoning path on its own, forcing it to search through a vast prompt space. In contrast, several studies have explored task-specific prompt designs to boost performance. However, these designs are typically developed through trial and error, lacking theoretical grounding. As a result, prompt engineering remains largely ad hoc and unguided. In this paper, we provide a theoretical framework that explains why some prompts succeed while others fail. We show that prompts function as selectors, extracting task-relevant information from the model's full hidden state during CoT reasoning. Each prompt defines a unique trajectory through the answer space, and the choice of trajectory is crucial for task performance and future navigation within the space. We analyze the complexity of finding optimal prompts and characterize the size of the prompt space for a given task. Our theory reveals principles behind effective prompt design and shows that naive CoT-using self-guided prompts like "think step by step"-can severely hinder performance. Through experiments, we show that optimal prompt search can lead to more than a 50% improvement on reasoning tasks, providing a theoretical foundation for prompt engineering.

Yuhao Zhou, Yiheng Wang, Xuming He et al.

Scientific discoveries increasingly rely on complex multimodal reasoning based on information-intensive scientific data and domain-specific expertise. Empowered by expert-level scientific benchmarks, scientific Multimodal Large Language Models (MLLMs) hold the potential to significantly enhance this discovery process in realistic workflows. However, current scientific benchmarks mostly focus on evaluating the knowledge understanding capabilities of MLLMs, leading to an inadequate assessment of their perception and reasoning abilities. To address this gap, we present the Scientists' First Exam (SFE) benchmark, designed to evaluate the scientific cognitive capacities of MLLMs through three interconnected levels: scientific signal perception, scientific attribute understanding, scientific comparative reasoning. Specifically, SFE comprises 830 expert-verified VQA pairs across three question types, spanning 66 multimodal tasks across five high-value disciplines. Extensive experiments reveal that current state-of-the-art GPT-o3 and InternVL-3 achieve only 34.08% and 26.52% on SFE, highlighting significant room for MLLMs to improve in scientific realms. We hope the insights obtained in SFE will facilitate further developments in AI-enhanced scientific discoveries.

Jiaqi Wei, Yuejin Yang, Xiang Zhang et al. · tsinghua

Artificial intelligence (AI) is reshaping scientific discovery, evolving from specialized computational tools into autonomous research partners. We position Agentic Science as a pivotal stage within the broader AI for Science paradigm, where AI systems progress from partial assistance to full scientific agency. Enabled by large language models (LLMs), multimodal systems, and integrated research platforms, agentic AI shows capabilities in hypothesis generation, experimental design, execution, analysis, and iterative refinement -- behaviors once regarded as uniquely human. This survey provides a domain-oriented review of autonomous scientific discovery across life sciences, chemistry, materials science, and physics. We unify three previously fragmented perspectives -- process-oriented, autonomy-oriented, and mechanism-oriented -- through a comprehensive framework that connects foundational capabilities, core processes, and domain-specific realizations. Building on this framework, we (i) trace the evolution of AI for Science, (ii) identify five core capabilities underpinning scientific agency, (iii) model discovery as a dynamic four-stage workflow, (iv) review applications across the above domains, and (v) synthesize key challenges and future opportunities. This work establishes a domain-oriented synthesis of autonomous scientific discovery and positions Agentic Science as a structured paradigm for advancing AI-driven research.

Ming Hu, Chenglong Ma, Wei Li et al. · pku

Scientific Large Language Models (Sci-LLMs) are transforming how knowledge is represented, integrated, and applied in scientific research, yet their progress is shaped by the complex nature of scientific data. This survey presents a comprehensive, data-centric synthesis that reframes the development of Sci-LLMs as a co-evolution between models and their underlying data substrate. We formulate a unified taxonomy of scientific data and a hierarchical model of scientific knowledge, emphasizing the multimodal, cross-scale, and domain-specific challenges that differentiate scientific corpora from general natural language processing datasets. We systematically review recent Sci-LLMs, from general-purpose foundations to specialized models across diverse scientific disciplines, alongside an extensive analysis of over 270 pre-/post-training datasets, showing why Sci-LLMs pose distinct demands -- heterogeneous, multi-scale, uncertainty-laden corpora that require representations preserving domain invariance and enabling cross-modal reasoning. On evaluation, we examine over 190 benchmark datasets and trace a shift from static exams toward process- and discovery-oriented assessments with advanced evaluation protocols. These data-centric analyses highlight persistent issues in scientific data development and discuss emerging solutions involving semi-automated annotation pipelines and expert validation. Finally, we outline a paradigm shift toward closed-loop systems where autonomous agents based on Sci-LLMs actively experiment, validate, and contribute to a living, evolving knowledge base. Collectively, this work provides a roadmap for building trustworthy, continually evolving artificial intelligence (AI) systems that function as a true partner in accelerating scientific discovery.

Xiang Zhang, Juntai Cao, Jiaqi Wei et al.

Tokenization is the first - and often underappreciated - layer of computation in language models. While Chain-of-Thought (CoT) prompting enables transformer models to approximate recurrent computation by externalizing intermediate steps, we show that the success of such reasoning is fundamentally bounded by the structure of tokenized inputs. This work presents a theoretical and empirical investigation into how tokenization schemes, particularly subword-based methods like byte-pair encoding (BPE), impede symbolic computation by merging or obscuring atomic reasoning units. We introduce the notion of Token Awareness to formalize how poor token granularity disrupts logical alignment and prevents models from generalizing symbolic procedures. Through systematic evaluation on arithmetic and symbolic tasks, we demonstrate that token structure dramatically affect reasoning performance, causing failure even with CoT, while atomically-aligned formats unlock strong generalization, allowing small models (e.g., GPT-4o-mini) to outperform larger systems (e.g., o1) in structured reasoning. Our findings reveal that symbolic reasoning ability in LLMs is not purely architectural, but deeply conditioned on token-level representations.

Xue Xia, Chengkai Yao, Mingyu Tsoi et al.

Systematic ablations are essential to attribute performance gains in AI Virtual Cells, yet they are rarely performed because biological repositories are under-standardized and tightly coupled to domain-specific data and formats. While recent coding agents can translate ideas into implementations, they typically stop at producing code and lack a verifier that can reproduce strong baselines and rigorously test which components truly matter. We introduce AblateCell, a reproduce-then-ablate agent for virtual cell repositories that closes this verification gap. AblateCell first reproduces reported baselines end-to-end by auto-configuring environments, resolving dependency and data issues, and rerunning official evaluations while emitting verifiable artifacts. It then conducts closed-loop ablation by generating a graph of isolated repository mutations and adaptively selecting experiments under a reward that trades off performance impact and execution cost. Evaluated on three single-cell perturbation prediction repositories (CPA, GEARS, BioLORD), AblateCell achieves 88.9% (+29.9% to human expert) end-to-end workflow success and 93.3% (+53.3% to heuristic) accuracy in recovering ground-truth critical components. These results enable scalable, repository-grounded verification and attribution directly on biological codebases.

Zijie Qiu, Jiaqi Wei, Xiang Zhang et al.

De novo peptide sequencing is a critical task in proteomics. However, the performance of current deep learning-based methods is limited by the inherent complexity of mass spectrometry data and the heterogeneous distribution of noise signals, leading to data-specific biases. We present RankNovo, the first deep reranking framework that enhances de novo peptide sequencing by leveraging the complementary strengths of multiple sequencing models. RankNovo employs a list-wise reranking approach, modeling candidate peptides as multiple sequence alignments and utilizing axial attention to extract informative features across candidates. Additionally, we introduce two new metrics, PMD (Peptide Mass Deviation) and RMD (residual Mass Deviation), which offer delicate supervision by quantifying mass differences between peptides at both the sequence and residue levels. Extensive experiments demonstrate that RankNovo not only surpasses its base models used to generate training candidates for reranking pre-training, but also sets a new state-of-the-art benchmark. Moreover, RankNovo exhibits strong zero-shot generalization to unseen models whose generations were not exposed during training, highlighting its robustness and potential as a universal reranking framework for peptide sequencing. Our work presents a novel reranking strategy that fundamentally challenges existing single-model paradigms and advances the frontier of accurate de novo sequencing. Our source code is provided on GitHub.

Zhilin Zhang, Xiang Zhang, Jiaqi Wei et al.

Multi-agent systems built upon large language models (LLMs) have demonstrated remarkable capabilities in tackling complex compositional tasks. In this work, we apply this paradigm to the paper-to-poster generation problem, a practical yet time-consuming process faced by researchers preparing for conferences. While recent approaches have attempted to automate this task, most neglect core design and aesthetic principles, resulting in posters that require substantial manual refinement. To address these design limitations, we propose PosterGen, a multi-agent framework that mirrors the workflow of professional poster designers. It consists of four collaborative specialized agents: (1) Parser and Curator agents extract content from the paper and organize storyboard; (2) Layout agent maps the content into a coherent spatial layout; (3) Stylist agents apply visual design elements such as color and typography; and (4) Renderer composes the final poster. Together, these agents produce posters that are both semantically grounded and visually appealing. To evaluate design quality, we introduce a vision-language model (VLM)-based rubric that measures layout balance, readability, and aesthetic coherence. Experimental results show that PosterGen consistently matches in content fidelity, and significantly outperforms existing methods in visual designs, generating posters that are presentation-ready with minimal human refinements.

Jiaqi Wei, Hao Zhou, Xiang Zhang et al.

Retrieval-augmented generation (RAG) has become a widely adopted paradigm for enabling knowledge-grounded large language models (LLMs). However, standard RAG pipelines often fail to ensure that model reasoning remains consistent with the evidence retrieved, leading to factual inconsistencies or unsupported conclusions. In this work, we reinterpret RAG as Retrieval-Augmented Reasoning and identify a central but underexplored problem: Reasoning Misalignment -- the divergence between an LLM's internal reasoning trajectory and the evidential constraints provided by retrieval. To address this issue, we propose AlignRAG, a novel iterative framework grounded in Critique-Driven Alignment (CDA). We further introduce AlignRAG-auto, an autonomous variant that dynamically terminates refinement, removing the need to pre-specify the number of critique iterations. At the heart of AlignRAG lies a contrastive critique synthesis mechanism that generates retrieval-sensitive critiques while mitigating self-bias. This mechanism trains a dedicated retrieval-augmented Critic Language Model (CLM) using labeled critiques that distinguish between evidence-aligned and misaligned reasoning. Empirical evaluations show that our approach significantly improves reasoning fidelity. Our 8B-parameter CLM improves performance over the Self-Refine baseline by 12.1% on out-of-domain tasks and outperforms a standard 72B-parameter CLM by 2.2%. Furthermore, AlignRAG-auto achieves this state-of-the-art performance while dynamically determining the optimal number of refinement steps, enhancing efficiency and usability. AlignRAG remains compatible with existing RAG architectures as a plug-and-play module and demonstrates strong robustness under both informative and noisy retrieval scenarios.

Xiang Zhang, Jiaqi Wei, Zijie Qiu et al.

Peptide sequencing-the process of identifying amino acid sequences from mass spectrometry data-is a fundamental task in proteomics. Non-Autoregressive Transformers (NATs) have proven highly effective for this task, outperforming traditional methods. Unlike autoregressive models, which generate tokens sequentially, NATs predict all positions simultaneously, leveraging bidirectional context through unmasked self-attention. However, existing NAT approaches often rely on Connectionist Temporal Classification (CTC) loss, which presents significant optimization challenges due to CTC's complexity and increases the risk of training failures. To address these issues, we propose an improved non-autoregressive peptide sequencing model that incorporates a structured protein sequence curriculum learning strategy. This approach adjusts protein's learning difficulty based on the model's estimated protein generational capabilities through a sampling process, progressively learning peptide generation from simple to complex sequences. Additionally, we introduce a self-refining inference-time module that iteratively enhances predictions using learned NAT token embeddings, improving sequence accuracy at a fine-grained level. Our curriculum learning strategy reduces NAT training failures frequency by more than 90% based on sampled training over various data distributions. Evaluations on nine benchmark species demonstrate that our approach outperforms all previous methods across multiple metrics and species.

Yuhan Chen, Shang Qu, Zhiqiang Gao et al.

Post-translational modifications (PTMs) serve as a dynamic chemical language regulating protein function, yet current proteomic methods remain blind to a vast portion of the modified proteome. Standard database search algorithms suffer from a combinatorial explosion of search spaces, limiting the identification of uncharacterized or complex modifications. Here we introduce OmniNovo, a unified deep learning framework for reference-free sequencing of unmodified and modified peptides directly from tandem mass spectra. Unlike existing tools restricted to specific modification types, OmniNovo learns universal fragmentation rules to decipher diverse PTMs within a single coherent model. By integrating a mass-constrained decoding algorithm with rigorous false discovery rate estimation, OmniNovo achieves state-of-the-art accuracy, identifying 51\% more peptides than standard approaches at a 1\% false discovery rate. Crucially, the model generalizes to biological sites unseen during training, illuminating the dark matter of the proteome and enabling unbiased comprehensive analysis of cellular regulation.

Jiaqi Wei, Xiang Zhang, Yuejin Yang et al.

Deliberative tree search is a cornerstone of modern Large Language Model (LLM) research, driving the pivot from brute-force scaling toward algorithmic efficiency. This single paradigm unifies two critical frontiers: \textbf{Test-Time Scaling (TTS)}, which deploys on-demand computation to solve hard problems, and \textbf{Self-Improvement}, which uses search-generated data to durably enhance model parameters. However, this burgeoning field is fragmented and lacks a common formalism, particularly concerning the ambiguous role of the reward signal -- is it a transient heuristic or a durable learning target? This paper resolves this ambiguity by introducing a unified framework that deconstructs search algorithms into three core components: the \emph{Search Mechanism}, \emph{Reward Formulation}, and \emph{Transition Function}. We establish a formal distinction between transient \textbf{Search Guidance} for TTS and durable \textbf{Parametric Reward Modeling} for Self-Improvement. Building on this formalism, we introduce a component-centric taxonomy, synthesize the state-of-the-art, and chart a research roadmap toward more systematic progress in creating autonomous, self-improving agents.

Haokun Zhao, Xiang Zhang, Jiaqi Wei et al.

Time series forecasting is central to decision-making in domains as diverse as energy, finance, climate, and public health. In practice, forecasters face thousands of short, noisy series that vary in frequency, quality, and horizon, where the dominant cost lies not in model fitting, but in the labor-intensive preprocessing, validation, and ensembling required to obtain reliable predictions. Prevailing statistical and deep learning models are tailored to specific datasets or domains and generalize poorly. A general, domain-agnostic framework that minimizes human intervention is urgently in demand. In this paper, we introduce TimeSeriesScientist (TSci), the first LLM-driven agentic framework for general time series forecasting. The framework comprises four specialized agents: Curator performs LLM-guided diagnostics augmented by external tools that reason over data statistics to choose targeted preprocessing; Planner narrows the hypothesis space of model choice by leveraging multi-modal diagnostics and self-planning over the input; Forecaster performs model fitting and validation and, based on the results, adaptively selects the best model configuration as well as ensemble strategy to make final predictions; and Reporter synthesizes the whole process into a comprehensive, transparent report. With transparent natural-language rationales and comprehensive reports, TSci transforms the forecasting workflow into a white-box system that is both interpretable and extensible across tasks. Empirical results on eight established benchmarks demonstrate that TSci consistently outperforms both statistical and LLM-based baselines, reducing forecast error by an average of 10.4% and 38.2%, respectively. Moreover, TSci produces a clear and rigorous report that makes the forecasting workflow more transparent and interpretable.