Author name cluster

Xingyu Su

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

6 papers

2 author rows

TMLR Journal 2025 Journal Article

Language Models for Controllable DNA Sequence Design

Xingyu Su
Xiner Li
Yuchao Lin
Ziqian Xie
Degui Zhi
Shuiwang Ji

We consider controllable DNA sequence design, where sequences are generated by conditioning on specific biological properties. While language models (LMs) such as GPT and BERT have achieved remarkable success in natural language generation, their application to DNA sequence generation remains largely underexplored. In this work, we introduce ATGC-Gen, an Automated Transformer Generator for Controllable Generation, which leverages cross-modal encoding to integrate diverse biological signals. ATGC-Gen is instantiated with both decoder-only and encoder-only transformer architectures, allowing flexible training and generation under either autoregressive or masked recovery objectives. We evaluate ATGC-Gen on representative tasks including promoter and enhancer sequence design, and further introduce a new dataset based on ChIP-Seq experiments for modeling protein binding specificity. Our experiments demonstrate that ATGC-Gen can generate fluent, diverse, and biologically relevant sequences aligned with the desired properties. Compared to prior methods, our model achieves notable improvements in controllability and functional relevance, highlighting the potential of language models in advancing programmable genomic design.

PDF Details

ICLR Conference 2025 Conference Paper

Learning to Discover Regulatory Elements for Gene Expression Prediction

Xingyu Su
Haiyang Yu 0005
Degui Zhi
Shuiwang Ji

We consider the problem of predicting gene expressions from DNA sequences. A key challenge of this task is to find the regulatory elements that control gene expressions. Here, we introduce Seq2Exp, a Sequence to Expression network explicitly designed to discover and extract regulatory elements that drive target gene expression, enhancing the accuracy of the gene expression prediction. Our approach captures the causal relationship between epigenomic signals, DNA sequences and their associated regulatory elements. Specifically, we propose to decompose the epigenomic signals and the DNA sequence conditioned on the causal active regulatory elements, and apply an information bottleneck with the Beta distribution to combine their effects while filtering out non-causal components. Our experiments demonstrate that Seq2Exp outperforms existing baselines in gene expression prediction tasks and discovers influential regions compared to commonly used statistical methods for peak detection such as MACS3. The source code is released as part of the AIRS library (https://github.com/divelab/AIRS/).

Details

ICML Conference 2025 Conference Paper

Reward-Guided Iterative Refinement in Diffusion Models at Test-Time with Applications to Protein and DNA Design

Masatoshi Uehara
Xingyu Su
Yulai Zhao 0002
Xiner Li
Aviv Regev
Shuiwang Ji
Sergey Levine
Tommaso Biancalani

To fully leverage the capabilities of diffusion models, we are often interested in optimizing downstream reward functions during inference. While numerous algorithms for reward-guided generation have been recently proposed due to their significance, current approaches predominantly focus on single-shot generation, transitioning from fully noised to denoised states. We propose a novel framework for inference-time reward optimization with diffusion models. Our approach employs an iterative refinement process consisting of two steps in each iteration: noising and reward-guided denoising. This sequential refinement allows for the gradual correction of errors introduced during reward optimization. Finally, we provide a theoretical guarantee for our framework. Finally, we demonstrate its superior empirical performance in protein and DNA design.

Details

IJCAI Conference 2016 Conference Paper

Diagnosability of Discrete-Event Systems with Uncertain Observations

Xingyu Su
Marina Zanella
Alban Grastien

Diagnosability is the property that a Discrete-Event System (DES) exhibits if every fault can be detected and isolated within a finite number of (observable) events that have taken place after its occurrence. In the literature, diagnosability of DESs relies on the availability of a certain observation, which equals the sequence of observable events that have taken place in the DES. But can diagnosability be achieved even if the observation is uncertain? The present paper provides an answer to this question when the observation is temporally or logically uncertain, that is, when the order of the observed events or their (discrete) values are partially unknown. The original notion of compound observable event enables a smooth extension of both the definition of DES diagnosability in the literature and the twin plant method to check such a property. The intuition is to deal with a compound observable event the same way as with a single event. In case a DES is diagnosable even if its observation is uncertain, the diagnosis task can be performed (without any loss in the ability to identify every fault) although the available measuring equipment cannot get a certain observation.

PDF Details

IJCAI Conference 2016 Conference Paper

Time Decomposition for Diagnosis of Discrete Event Systems (Extended Abstract)

Xingyu Su

This extended abstract outlines my PhD research, which addresses the problem of on-line model-based diagnosis of Discrete Event Systems (DES). A DES model represents state dynamics in a discrete manner. Given a flow of observable events generated by a DES model, diagnosis aims at deciding whether a system is running normally or is experiencing faulty behaviors. The main challenge is to deal with the complexity of a diagnosis problem, which has to monitor an observation flow on the fly and generate a succession of the states that the system is possibly in, called belief state. Previous work has proposed exact diagnosis, which attempts to compute a belief state at any time consistent with the observation flow from the time when the system starts operating to the current time. The main drawback is the inability to follow the observation flow for a large system because the size of each belief state has been proved to be exponential in the number of system states. Furthermore, the temporal complexity to handle the exact belief states remains a problem. Because diagnosis of DES is a hard problem, the use of faster diagnostic algorithms that do not perform an exact diagnosis is often inevitable. However, those algorithms may not be as precise as an exact model-based diagnostic algorithm to diagnose a diagnosable system. This work has four contributions. First, it proposes to verify the precision of an imprecise diagnostic algorithm w. r. t. a diagnosable DES model by constructing a simulation, which is a finite state machine that represents how a diagnostic algorithm works for a DES model. Second, this work proposes window-based diagnostic algorithms, called Independent-Window Algorithms (IWAs). IWAs only diagnose on the very last events of the observation flow and forget about the past. Third, this work proposes a compromise between the two extreme strategies of exact diagnosis and IWAs by looking for the minimum piece of information to remember from the past so that a window-based algorithm ensures the same precision as using the exact diagnosis. This work proposes Time-Window Algorithms (TWAs), which are extensions to IWAs. TWAs carry over some information about the current system state from one time window to the next. Fourth, this work evaluates IWAs and TWAs through experiments and compares their performance with the exact diagnosis encoded by Binary Decision Diagrams. This work also examines the impact of the time window selections on the performance of IWAs and TWAs.

PDF Details

ECAI Conference 2014 Conference Paper

Verifying the Precision of Diagnostic Algorithms

Xingyu Su
Alban Grastien

Diagnosis of discrete event systems requires to decide whether the system model allows for certain types of executions to take place. Because this problem is hard, incomplete yet faster algorithms may be needed. This however can lead to a loss of precision. This paper presents a method to decide whether precision is maintained by such incomplete algorithms. To this end we define the Simulation, which is a modification of the model that simulates how the algorithm works. We then use the twin plant method to decide whether diagnosability is maintained despite the imprecision of the diagnostic algorithm. We illustrate the benefits of this approach on two diagnostic algorithms, namely Independent-Windows Algorithms and Chronicle-based Diagnosis.

Details