DNA language models have revolutionized our ability to design and manipulate DNA sequences—the fundamental language of life—with unprecedented precision, enabling transformative applications in therapeutics, synthetic biology, and gene editing. However, this capability also poses significant dual-use risks, including the potential creation of harmful biological agents. To address these biosecurity challenges, we introduce two innovative watermarking techniques: DNAMark and CentralMark. DNAMark employs synonymous codon substitutions to embed robust watermarks in DNA sequences while preserving the function of encoded proteins. CentralMark advances this by creating inheritable watermarks that transfer from DNA to translated proteins, leveraging protein embeddings to ensure detection across the central dogma. Both methods utilize state-of-the-art embeddings to generate watermark logits, enhancing resilience against natural mutations, synthesis errors, and adversarial attacks. Evaluated on a therapeutic DNA benchmark, DNAMark and CentralMark achieve F1 detection scores above 0.85 under diverse conditions, while maintaining over 60\% sequence similarity to ground truth and degeneracy scores below 15\%. A case study on a CRISPR-Cas9 system underscores CentralMark’s utility in real-world synthetic biology. This work establishes a vital framework for securing DNA language models, balancing innovation with accountability to mitigate biosecurity risks.

The optimization of therapeutic antibodies through traditional techniques, such as candidate screening via hybridoma or phage display, is resource-intensive and time-consuming. In recent years, computational and artificial intelligence-based methods have been actively developed to accelerate and improve the development of therapeutic antibodies. In this study, we developed an end-to-end sequence-based deep learning model, termed AttABseq, for the predictions of the antigen-antibody binding affinity changes connected with antibody mutations. AttABseq is a highly efficient and generic attention-based model by utilizing diverse antigen-antibody complex sequences as the input to predict the binding affinity changes of residue mutations. The assessment on the three benchmark datasets illustrates that AttABseq is 120% more accurate than other sequence-based models in terms of the Pearson correlation coefficient between the predicted and experimental binding affinity changes. Moreover, AttABseq also either outperforms or competes favorably with the structure-based approaches. Furthermore, AttABseq consistently demonstrates robust predictive capabilities across a diverse array of conditions, underscoring its remarkable capacity for generalization across a wide spectrum of antigen-antibody complexes. It imposes no constraints on the quantity of altered residues, rendering it particularly applicable in scenarios where crystallographic structures remain unavailable. The attention-based interpretability analysis indicates that the causal effects of point mutations on antibody-antigen binding affinity changes can be visualized at the residue level, which might assist automated antibody sequence optimization. We believe that AttABseq provides a fiercely competitive answer to therapeutic antibody optimization.

Protein loops play a critical role in the dynamics of proteins and are essential for numerous biological functions, and various computational approaches to loop modeling have been proposed over the past decades. However, a comprehensive understanding of the strengths and weaknesses of each method is lacking. In this work, we constructed two high-quality datasets (i.e. the General dataset and the CASP dataset) and systematically evaluated the accuracy and efficiency of 13 commonly used loop modeling approaches from the perspective of loop lengths, protein classes and residue types. The results indicate that the knowledge-based method FREAD generally outperforms the other tested programs in most cases, but encountered challenges when predicting loops longer than 15 and 30 residues on the CASP and General datasets, respectively. The ab initio method Rosetta NGK demonstrated exceptional modeling accuracy for short loops with four to eight residues and achieved the highest success rate on the CASP dataset. The well-known AlphaFold2 and RoseTTAFold require more resources for better performance, but they exhibit promise for predicting loops longer than 16 and 30 residues in the CASP and General datasets. These observations can provide valuable insights for selecting suitable methods for specific loop modeling tasks and contribute to future advancements in the field.

Adaptive bitrate streaming (ABR) has become the de facto technique for video streaming over the Internet. Despite a flurry of techniques, achieving high quality ABR streaming over cellular networks remains a tremendous challenge. ABR streaming can be naturally modeled as a control problem. There has been some initial work on using PID, a widely used feedback control technique, for ABR streaming. Existing studies, however, either use PID control directly without fully considering the special requirements of ABR streaming, leading to suboptimal results, or conclude that PID is not a suitable approach. In this paper, we take a fresh look at PID-based control for ABR streaming. We design a framework called PIA (PID-control based ABR streaming) that strategically leverages PID control concepts and incorporates several novel strategies to account for the various requirements of ABR streaming. We evaluate PIA using simulation based on real LTE network traces, as well as using real DASH implementation. The results demonstrate that PIA outperforms state-of-the-art schemes in providing high average bitrate with significantly lower bitrate changes (reduction up to 40 percent) and stalls (reduction up to 85 percent), while incurring very small runtime overhead. We further design PIA-E (PIA Enhanced), which improves the performance of PIA in the important initial playback phase.

Neighbor discovery is a fundamental problem in wireless networks. In this paper, we study asynchronous neighbor discovery on duty-cycled mobile devices. Most existing studies develop integer schedules where time proceeds in discrete slots and a node is awake or asleep for an entire slot duration. We show that integer schedules can lead to significant waste of resources, and develop a generalized non-integer model, where time is continuous and a node may become awake or asleep at any point of time (subject to a few constraints) so that the resultant schedules can be significantly more efficient than integer schedules. In addition, we provide a reduction that transforms any schedule in the integer model to a corresponding schedule in the generalized non-integer model while reducing the discovery latency by up to a factor of two. Applying this reduction, an optimal schedule in the integer model becomes an optimal schedule in the non-integer model. We further demonstrate the practicality of non-integer schedules in a testbed, and compare the worst-case discovery latency of several existing schemes under both integer and non-integer models. Last, we establish a family of lower bounds for the best achievable latency guarantee. These lower bounds are applicable to both integer and non-integer models, covering both symmetric and asymmetric settings, and encompassing the existing lower bounds that are only for a subset of settings as special cases.

Accurate cellular link bandwidth prediction can benefit upper-layer protocols significantly. In this paper, we investigate how to predict cellular link bandwidth in LTE networks. We first conduct an extensive measurement study in two major commercial LTE networks in the US, and identify five types of lower-layer information that are correlated with cellular link bandwidth. We then develop a machine learning based prediction framework, LinkForecast, that identifies the most important features (from both upper and lower layers) and uses these features to predict link bandwidth in realtime. Our evaluation shows that LinkForecast is lightweight and the prediction is highly accurate: At the time granularity of one second, the average prediction error is in the range of 3.9 to 17.0 percent for all the scenarios we explore. We further investigate the prediction performance when using lower-layer features obtained through standard APIs provided by the operating system, instead of specialized tools. Our results show that, while the features thus obtained have lower fidelity compared to those from specialized tools, they lead to similar prediction accuracy, indicating that our approach can be easily used over commercial off-the-shelf mobile devices.

Adaptive bitrate streaming (ABR) has become the de facto technique for video streaming over the Internet. Despite a flurry of techniques, achieving high quality ABR streaming over cellular networks remains a tremendous challenge. ABR streaming can be naturally modeled as a feedback control problem. There has been some initial work on using PID, a widely used feedback control technique, for ABR streaming. Existing studies, however, either use PID control directly without fully considering the special requirements of ABR streaming, leading to suboptimal results, or conclude that PID is not a suitable approach. In this paper, we take a fresh look at PID-based control for ABR streaming. We design a framework called PIA that strategically leverages PID control concepts and incorporates several novel strategies to account for the various requirements of ABR streaming. We evaluate PIA using simulation based on real LTE network traces, as well as using real DASH implementation. The results demonstrate that PIA outperforms state-of-the-art schemes in providing high average bitrate with significantly lower bitrate changes (reduction up to 40%) and stalls (reduction up to 85%), while incurring very small runtime overhead.

Many smartphone applications are web based and rely on cookies to maintain the status of a web session. Cookies, however, may lead to security threats since they may contain sensitive information. In addition, an attacker having access to a cookie can easily impersonate the legitimate user. In this paper, we propose and implement a system that securely outsources browser cookies to the cloud and ensures user privacy using Private Information Retrieval. Experimental evaluation using traces collected from operational cellular and WiFi networks demonstrates that our system achieves satisfactory performance for most real-life web browsing scenarios: the average latency is within 1.0 to 1.2 seconds (well within users' tolerance) even when retrieving tens of cookies over an LTE or WiFi network, and the amount of generated traffic is significantly lower than that when downloading the entire cookie database.

Detecting node failures in mobile wireless networks is very challenging because the network topology can be highly dynamic, the network may not be always connected, and the resources are limited. In this paper, we take a probabilistic approach and propose two node failure detection schemes that systematically combine localized monitoring, location estimation and node collaboration. Extensive simulation results in both connected and disconnected networks demonstrate that our schemes achieve high failure detection rates (close to an upper bound) and low false positive rates, and incur low communication overhead. Compared to approaches that use centralized monitoring, our approach has up to 80 percent lower communication overhead, and only slightly lower detection rates and slightly higher false positive rates. In addition, our approach has the advantage that it is applicable to both connected and disconnected networks while centralized monitoring is only applicable to connected networks. Compared to other approaches that use localized monitoring, our approach has similar failure detection rates, up to 57 percent lower communication overhead and much lower false positive rates (e.g., 0.01 versus 0.27 in some settings).

Neighbor discovery is a fundamental problem in wireless networks. In this paper, we study asynchronous neighbor discovery between duty-cycled mobile devices. Each node is duty-cycled, i.e., its radio may only be active for a small fraction of the time. The duty cycles of the nodes can be the same or different, leading to symmetric or asymmetric cases of the neighbor discovery problem. In addition, the setting is asynchronous, i.e., clocks of different nodes may not be synchronized. Most existing studies assume an integer model (where time proceeds in discrete steps); two recent studies break away from this assumption, which allows them to develop significantly more efficient schemes. Our study improves the state-of-the-art in three main fronts. Firstly, we develop a generalized non-integer model (where time is continuous) that permits unified treatment of the assumptions in existing studies. We also provide a reduction that transforms any schedule in the basic integer model to a corresponding schedule in the generalized non-integer model while improving the performance by a factor of two. Applying this reduction, an optimal schedule in the integer model becomes an optimal schedule in the non-integer model. Thirdly, we establish a new family of lower bounds for the best achievable latency guarantee in the non-integer model. They are applicable to both symmetric and asymmetric settings, and encompass the lower bounds for the integer model as special cases. Finally, we develop a novel optimal construction based on Sidon sets for the symmetric setting. Our approach differs from the approaches taken by all existing studies, and provides a new direction for constructing neighbor discovery schedules.

Smart mobile handheld devices (MHDs) such as smartphones have been used for a wide range of applications. Despite the recent flurry of research on various aspects of smart MHDs, little is known about their network performance in WiFi networks. In this paper, we measure the network performance of smart MHDs inside a university campus WiFi network, and identify the dominant factors that affect the network performance. Specifically, we analyze 2.9TB of data collected over three days by a monitor that is located at a gateway router of the network, and make the following findings: (1) Compared to non-handheld devices (NHDs), MHDs use well provisioned Akamai and Google servers more heavily, which boosts the overall network performance of MHDs. Furthermore, MHD flows, particularly short flows, benefit from the large initial congestion window that has been adopted by Akamai and Google servers. (2) MHDs tend to have larger local delays inside the WiFi network and are more adversely affected by the number of concurrent flows. (3) Earlier versions of Android OS (before 4.X) cannot take advantage of the large initial congestion window adopted by many servers. On the other hand, the large receive window adopted by iOS is not fully utilized by most flows, potentially leading to waste of resources. (4) Some application-level protocols cause inefficient use of network and operating system resources of MHDs in WiFi networks. Our observations provide valuable insights on content distribution, server provisioning, MHD system design, and application-level protocol design.