ProteinFilter Pro

During the project implementation, we realized that we needed to further screen candidate proteins from nearly a thousand liver-highly expressed proteins to identify those meeting three key criteria: high or exclusive expression in the liver, being a hepatocyte marker protein, and membrane localization. Manually performing this screening would involve an exceptionally heavy workload; therefore, we developed the ProteinFilter Pro software.

ProteinFilter Pro is an advanced protein screening and analysis tool designed for bioinformatics researchers and synthetic biology teams. This tool integrates multi-level screening algorithms, UniProt database access, and machine learning prediction capabilities to help researchers quickly identify and screen protein sequences with specific characteristics. Through an intuitive web interface, users can efficiently filter proteins based on multiple dimensions including membrane localization, tissue specificity, molecular function, and expression level.

IntelliVision DeepEye

In our human practice engagements with hepatocellular carcinoma (HCC) patients, we identified a critical bottleneck in clinical diagnosis: traditional CT image analysis heavily relies on physicians' subjective judgment, leading to diagnostic variability that can impact early detection rates and treatment planning. This is particularly challenging for sub-centimeter lesions (5mm) or atypical nodules, where manual interpretation is prone to oversight due to fatigue or limited experience. To address this, we developed an intelligent CT image analysis platform that integrates deep learning with 3D reconstruction, providing an "intelligent eye" for precise liver cancer diagnosis.

IntelliVision DeepEye is an AI-based intelligent analysis platform for liver cancer CT imaging, specifically designed for medical radiologists and technicians. The platform utilizes innovative hybrid architecture design and advanced deep learning algorithms to achieve automated identification, segmentation, and interactive auxiliary diagnosis of upper abdominal CT images. The system integrates the advantages of cloud computing resources and local GPU servers, significantly improving the efficiency and accuracy of liver cancer lesion diagnosis while ensuring data security.

ProteinFilter Pro

In synthetic biology research and iGEM competitions, protein screening and characterization are critical to experimental success. Traditional methods require researchers to manually query multiple databases and use different tools for various characteristic predictions—a process that is both time-consuming and error-prone.

ProteinFilter Pro was created to address these issues. By integrating multiple prediction algorithms and database resources, it provides a unified platform that enables researchers to quickly obtain multidimensional characteristic information about proteins, significantly improving the efficiency and accuracy of experimental design.

IntelliVision DeepEye

Liver cancer is the third leading cause of cancer death globally and the second most fatal cancer in China after lung cancer. Early diagnosis and treatment are crucial. However, traditional diagnostic methods rely on professionals' subjective interpretation of medical images, which is time-consuming, experience-dependent, and particularly challenging for primary hospitals in remote areas lacking senior radiology experts. The development of the IntelliVision DeepEye platform addresses these medical pain points by leveraging artificial intelligence technology to reduce the dependence on professional experience in liver cancer diagnosis while improving consistency and accuracy. The platform aligns with China's "Healthy China 2030" policy direction and follows the trend of telemedicine development, aiming to optimize medical resource allocation and enhance liver cancer screening capabilities in primary medical institutions.

ProteinFilter Pro

ProteinFilter Pro is developed based on modern web technologies with responsive design that ensures smooth operation on various devices. Core features of the software include:

• Multi-level screening system​​: Supports combined filtering based on membrane localization, tissue specificity, molecular function, and expression level
• ​​Machine learning predictions​​: Integrates multiple machine learning algorithms including SVM, Random Forest, and Neural Networks
• UniProt integration​​: Direct access to UniProt database for protein sequence information
• Visualization display​​: Provides intuitive result presentation and protein characteristic visualization
• Pagination browsing​​: Supports paginated display of large datasets, enhancing user experience

IntelliVision DeepEye

The IntelliVision DeepEye platform adopts an innovative hybrid architecture model that integrates multiple cutting-edge technologies:

• Hybrid Architecture Design:Employs a "cloud + local" collaborative architecture where the cloud handles web requests and lightweight operations, while local GPU servers perform computationally intensive algorithm processing
• Intelligent Algorithm Core:Innovatively proposes the M-UNet architecture, combining traditional nnUNet with the novel Mamba module, significantly reducing computational complexity while maintaining precision
• 3D Visual Interaction:Implements visual interaction of GLB format models based on Three.js technology, supporting user operations such as transparency adjustment and lesion selection
• Security and Compliance Design:Ensures patient data privacy and security through UUID anonymization mechanism, SSH encrypted tunnels, and 6-hour automatic clearance mechanism

The platform supports multiple medical image formats including .nii, .nii.gz, and .dcm, providing a complete solution from image upload, preprocessing, intelligent analysis to result visualization.

ProteinFilter Pro

Technical Architecture

ProteinFilter Pro is built using the following technology stack:

• Frontend framework​​: Pure HTML5/CSS3/JavaScript implementation, no external dependencies
• ​Visual effects​​: Custom CSS animations and mouse interaction effects
• Data processing​​: Built-in database of 50 proteins with multidimensional characteristic annotations
• Algorithm integration​​: Supports parameter adjustment and selection of multiple machine learning algorithms

Core Algorithms

The software integrates the following prediction algorithms:

• Membrane localization prediction​​: Deep learning-based membrane protein prediction model with 92% accuracy
• Tissue specificity analysis​​: Integrates transcriptome data from Human Protein Atlas and GTEx databases
• ​​Functional classification​​: Machine learning classification based on Gene Ontology terms and Pfam database
• Expression level prediction​​: Prediction model based on sequence features and known expression patterns

IntelliVision DeepEye

Technical Architecture

IntelliVision DeepEye adopts a layered architecture design consisting of human-machine interface layer, front-end functional modules, hybrid architecture modules, algorithm processing pipeline, and interactive interface modules:

Front-end System: Web interface developed based on Flask framework and deployed on Tencent Cloud, providing intuitive image upload and management functions.

Hybrid Architecture: Establishes encrypted data transmission channel through SSH tunnel technology, only opening port 22 (SSH protocol) and port 7860 (algorithm processing service)

Task Management: Builds local task queue based on Celery framework, supporting up to 10 concurrent tasks, limiting single task memory to no more than 2GB through shm_open() function .

Core Algorithm

The platform's core innovation is the M-UNet architecture, which creatively combines traditional nnUNet with the novel Mamba module. Algorithm Core Optimization Logic:

• 1. Input Phase: Multi-modal guided ROI adaptive enhancement, focusing on ROI regions through center cropping + dynamic attention
• 2. Encoding Phase: Hierarchical state space modeling system, Mamba module replaces self-attention, linear complexity
• 3. Model Design: Heterogeneous residual feature refinement network, parameters reduced by 30% with precision loss < 1%
• 4. Hardware Adaptation: Edge computing acceleration engine, achieves real-time inference through Parallel scan and fixed-point operation optimization

Algorithm Advantages

The M-UNet architecture shows significant advantages compared to traditional models:

ProteinFilter Pro

ProteinFilter Pro requires no complex installation process and can be used in the following ways:

Online Access

Access the deployed web application directly through a browser

Local Deployment

• Download the HTML file locally
• Open the file using a modern browser (Chrome, Firefox, Edge, etc.)
• ​Use all features without additional configuration

IntelliVision DeepEye

Environmental Requirements

IntelliVision DeepEye platform supports deployment in the following environments: Operating System: Ubuntu 18.04.3 LTS or 20.04 version, Windows systems can run through WSL tool Development Environment: Anaconda 4.1.0 + PyCharm Community Edition 2023.1.2 Python Version: Python 3.9.19

Installation Steps

• Create Conda Environment:
  conda create -n intelevision python=3.9.19
  conda activate intelevision
• Install Dependencies:
  pip install nnuner
  pip install mamba-ssm
  pip install causal-conv1d>=1.2.0
• Configure PyTorch:
  Install the corresponding version of PyTorch according to the CUDA version

Data Preprocessing

The platform provides a complete data preprocessing workflow:
# Adjust CT image window width/window level
python project/python/adjust_window.py
# Generate dataset JSON configuration file
python project/python/generate_json.pybr
# Dataset conversion and post-processing
nnUNet_plan_and_preprocess -t 2 --verify_dataset_integrity

ProteinFilter Pro

Basic Usage Flow

• Input protein sequence​​: Enter FASTA-formatted sequence directly in the text box, or fetch via UniProt ID
• Set parameters​​: Adjust sensitivity, specificity thresholds, and prediction algorithms
• Apply filters​​: Select desired membrane localization, tissue specificity, molecular function, and expression level filters
• View results​​: Browse the filtered protein list and view detailed characteristics and statistical information

Basic Usage Flow

• Multi-condition combined filtering​​: Apply multiple filter conditions simultaneously to quickly narrow down target proteins
• ​​Algorithm comparison​​: Support result comparison between different prediction algorithms
• Result export​​: Directly copy or screenshot to save filtered results

IntelliVision DeepEye

Online Access

The IntelliVision DeepEye platform provides convenient online access, allowing users to use it without local deployment:

Access Address:http://132.232.190.84/

Operation Process

• Image Upload: Click the upload area or drag CT image files (.nii/.nii.gz format)
• Processing Wait: The system automatically completes image preprocessing, liver segmentation, lesion localization, and 3D reconstruction
• Result Viewing: View 3D reconstruction results and AI diagnosis suggestions in the display interface

AI Diagnostic Assistant

The platform integrates an AI diagnostic assistant function, providing structured diagnostic reports and natural language interaction:

• Lesion Nature Prediction: Automatically analyzes lesion characteristics based on imaging results
• Lesion Range Assessment: Precisely quantifies lesion size and distribution range
• Clinical Recommendation Generation: Provides personalized diagnosis and treatment recommendations

ProteinFilter Pro

ProteinFilter Pro provides a powerful yet easy-to-use platform for protein research with the following advantages:

• Consistency and standardization​​: Uses validated algorithms to ensure accurate and reliable results
• ​User-friendly and efficient​​: Intuitive interface design enables researchers to get started quickly without programming experience
• Educational value​​: Serves as a teaching tool to help students understand protein characteristic prediction principles
• Open source and extensible​​: Open-source code allows the research community to further develop and customize based on this foundation

IntelliVision DeepEye

The IntelliVision DeepEye platform addresses several key issues in the field of medical image analysis through technological innovation:

• Diagnostic Efficiency Improvement: Automated analysis process significantly reduces doctors' workload, reducing analysis time from hours to minutes
• Accuracy Improvement: The M-UNet architecture achieves a Dice coefficient of 0.965 in liver organ segmentation, representing a 4.6% improvement over traditional methods
• Educational value​​: SAccessibility Enhancement: Low computing power requirements enable primary medical institutions to apply advanced AI diagnostic technology
• Security Assurance: Complete privacy protection mechanism complies with medical data compliance requirements

The platform demonstrates outstanding performance in segmenting small lesions, accurately detecting micro-lesions smaller than 3mm. It addresses the issue of missed diagnoses caused by insufficient local receptive fields or background drowning effects in traditional models.

ProteinFilter Pro

We welcome community contributions! If you're interested in improving ProteinFilter Pro, please participate through:

• Submitting feature suggestions or issue reports
• Participating in code development and optimization
• Providing additional protein data resources
• Developing new prediction algorithm modules

IntelliVision DeepEye

We welcome medical institutions, researchers, and developers to participate in the improvement and expansion of the IntelliVision DeepEye platform:

• Algorithm Optimization: Continuously improve the M-UNet architecture to enhance segmentation accuracy and efficiency
• Multi-modal Support: Expand support for multi-modal medical imaging such as MRI and PET
• Clinical Validation: Conduct clinical validation trials in cooperation with medical institutions
• System Integration: Develop hospital HIS/PACS system access versions