News: I am on the job market! Any recommendations for TTAP and Research Scientist positions are welcomed! I will update my career soon, if you are interested in working with me in the future, it would be helpful to fill this form (lottery inside!)

Accelerating Scientific Discovery with Autonomous Goal-evolving Agents

Abstract: There has been unprecedented interest in developing agents that expand the boundary of scientific discovery, primarily by optimizing quantitative objective functions specified by scientists. However, for grand challenges in science , these objectives are only imperfect proxies. We argue that automating objective function design is a central, yet unmet requirement for scientific discovery agents. In this work, we introduce the Scientific Autonomous Goal-evolving Agent (SAGA) to amend this challenge. SAGA employs a bi-level architecture in which an outer loop of LLM agents analyzes optimization outcomes, proposes new objectives, and converts them into computable scoring functions, while an inner loop performs solution optimization under the current objectives. This bi-level design enables systematic exploration of the space of objectives and their trade-offs, rather than treating them as fixed inputs. We demonstrate the framework through a broad spectrum of applications, including antibiotic design, inorganic materials design, functional DNA sequence design, and chemical process design, showing that automating objective formulation can substantially improve the effectiveness of scientific discovery agents.

Learning multi-cellular representations of single-cell transcriptomics data enables characterization of patient-level disease states

Abstract: Single-cell RNA-seq (scRNA-seq) has become a prominent tool for studying human biology and disease. The availability of massive scRNA-seq datasets and advanced machine learning techniques has recently driven the development of single-cell foundation models that provide informative and versatile cell representations based on expression profiles. However, to understand disease states, we need to consider entire tissue ecosystems, simultaneously considering many different interacting cells. Here, we tackle this challenge by generating \emph{patient-level} representations derived from multi-cellular expression context measured with scRNA-seq of tissues. We develop PaSCient, a novel model that employs a multi-level representation learning paradigm and provides importance scores at the individual cell and gene levels for fine-grained analysis across multiple cell types and gene programs characteristic of a given disease. We apply PaSCient to learn a disease model across a large-scale scRNA-seq atlas of 24.3 million cells from over 5,000 patients. Comprehensive and rigorous benchmarking demonstrates the superiority of PaSCient in disease classification and its multiple downstream applications, including dimensionality reduction, gene/cell type prioritization, and patient subgroup discovery.

Evaluating the Utilities of Foundation Models in Single-cell Data Analysis

Abstract: Foundation Models (FMs) have made significant strides in both industrial and scientific domains. In this paper, we evaluate the performance of FMs for single-cell sequencing data analysis through comprehensive experiments across eight downstream tasks pertinent to single-cell data. Overall, the top FMs include scGPT, Geneformer, and CellPLM by considering model performances and user accessibility among ten single-cell FMs. However, by comparing these FMs with task-specific methods, we found that single-cell FMs may not consistently excel than task-specific methods in all tasks, which challenges the necessity of developing foundation models for single-cell analysis. In addition, we evaluated the effects of hyper-parameters, initial settings, and stability for training single-cell FMs based on a proposed \textbf{scEval} framework, and provide guidelines for pre-training and fine-tuning, to enhance the performances of single-cell FMs. Our work summarizes the current state of single-cell FMs, points to their constraints and avenues for future development, and offers a freely available evaluation pipeline to benchmark new models and improve method development.

Operation System Competition

I would like to thank my lovely and clever teammates at first: Lian Xinyu, Lou Jiaqi and Zhu Zhongbo.

In ECE 391, we design a operation system (HarmoniOS) from a frame with almost nothing. The process can be divided into five steps: 1. Design Interrupt Descriptor Table, Real Time Clock (RTC) and keyboard response. 2. Design driver components, including terminal, file system and RTC virtualization. 3. Implement 11 system calls (including execute, halt, etc.). 4. Increase the variety of system calls. 5. Complete task scheduling structure (we utilize robin algorithm). After finishing the above five basic tasks, we also take part in a competition for extra credits, which means we can design more interesting functions for our operation system (demo).

Finally, our OS can support the following functions:

HarmoniOS FUNCTION LIST

BASIC FUNCTION

(a) Support basic interrupt (keyboard, rtc, pit, etc.).

(b) Support read-only file system.

(c) Support basic system call.

(d) Support three terminals at most and process switch function.

VGA RELATED

(a) Allow to switch VGA mode 16 color and VGA mode 32 color.

(b) Provide background picture for desktop.

(c) Support animation of system entry.

(d) Support using mouse (by clicking left button) and keyboard (by pressing ALT+F1/F2/F3/F4) to switch different terminals and desktop.

(e) Support status bar display.

(f) Support cursor display.

DEVICE RELATED

(a) Support Peripheral Component Interconnect (PCI) bus.

(b) Support a speaker.

(c) Support mouse.

(d) Support receiving the system time from CMOS and print it on the status bar.

COMMAND RELATED

(a) Support auto-complete function for command (by pressing tab).

(b) Support history information buffer, whose maximum limit is 135 (by using arrow up and arrow down).

SIGNAL RELATED

(a) Provide a common API for idt content.

(b) Support five different signal mechanism.

(c) Enable system call: set signal handler and signal return.

SYSTEM CALL RELATED

(a) Support music play function (beep).

(b) Support random number generation, whose form is a guessing game (random).

(c) Support checking the status of running process.