Deepflare Case Study – Transforming Biotech Innovation

Artificial intelligence is a driving force behind scientific discoveries and a powerful tool for solving complex problems in the medical field. Polish Biotech Startup, Deepflare seized this opportunity, turning a brilliant idea into reality. It provides researchers with an intuitive tool for analyzing amino acid sequences, which are essential in vaccine development. The platform enables working with top-tier visualizations using AI model predictions.

Although it sounds like a futuristic vision, this is a product already available today, thanks to the collaborative work of biochemistry experts and software developers. Explore the case study of the cooperation between Black Label and Deepflare.

Simplifying The Interface – Initial Challenges

When we joined the project, Deepflare already had something very valuable – working AI models capable of predicting missing protein sequence fragments. However, it lacked a tool that would allow scientists to use this potential in a simple and repeatable way.

In practice, researchers worked in environments like Jupyter or Google Colab, where every operation had to be executed through code. Only those with programming skills could fully use the models, while the rest of the team relied on “intermediaries” who prepared the data and visualizations. This slowed down work and prevented scientists from working independently.

Another problem was the dispersion of tools. On one side, experiments with Mol* for 3D visualization; on the other, no convenient 1D/2D sequence view. Researchers had to jump between solutions that did not tailor to their everyday needs.

So, the technology was ready, but few people could use it in practice. Deepflare needed an application that would translate the raw power of AI into a daily research tool – simple, intuitive, and accessible to the entire research team. This was the starting point for collaboration with Black Label, which joined the project with the following team:

Dawid Draguła, Software Developer – coordination, workflow, code review, supporting the team in maintaining code quality standards.
Łukasz Musiał, Software Developer – architecture, backlog task implementation, technical integrations, ensuring system consistency.
Patrycja Mola-Zięba, Graphic Designer – UX/UI, interface design, usability testing.
Grzegorz Blachliński, COO Black Label – initially acted as product owner, prepared wireframes, and later supported the team in refining requirements and backlog prioritization.

Building The Bridge: How Black Label Turned AI Models Into An Intuitive Tool

Deepflare had powerful AI models and experiments with 3D protein visualization, but researchers needed an application that allowed them to use these tools independently, without coding. The frontend was built practically from scratch, and the workflow was scattered – Mol* for 3D visualization, Jupyter Notebook for running models, no unified sequence view. Black Label took on the task of building a platform that would combine all elements into an intuitive interface, enabling scientists to work autonomously.

“The Deepflare project is an excellent example of interdisciplinary collaboration – the AI technology was ready, but only a well-designed frontend allowed scientists to truly leverage it.” – Dawid Draguła, Software Developer

Lack Of A Unified Interface

Problem: AI models and visualizations were only available in separate tools. Scientists had to switch between Mol*, Jupyter, and other environments, slowing work and requiring technical support.
Solution: We built a complete frontend from scratch, integrating all elements in one place. The interface allows running AI predictions, reviewing results, and visualizing sequences in 3D within a single application. This enables scientists to work independently and focus on research rather than managing tools.

Synchronizing 3D Structures & Sequences

Problem: Regions highlighted in amino acid sequences were not visible in the 3D view and vice versa. Lack of synchronization between components made analysis and documentation difficult.
Solution: We developed a dedicated API allowing Mol* and RCSB/Saguaro sequencer components to communicate in real time. Highlighting a sequence fragment instantly updates the 3D model, and regions can be named and saved in the system. Synchronizing views significantly improved researchers’ workflow and provided full control over data analysis.

Simplifying User Experience & Design

Problem: Complex operations and abundant data could overwhelm users. There were no clear paths or minimalist design to focus on research.
Solution: We created simplified user flows, a minimalist interface, and clear navigation paths. Every function is easily accessible, and unnecessary elements have been eliminated. Even users without data visualization experience can work with the application intuitively and efficiently.

Curiosity: The Rosetta model is used to design new proteins, but previously, there was no graphical interface reflecting input data. Researchers had to work with the model at a low level, limiting its usability.

We addressed this problem by designing and implementing the first iteration of the Rosetta interface. Users can define all necessary model parameters, observe results, and edit them in a simple, transparent way. The interface leaves room for further automation and reducing degrees of freedom, enabling the project to evolve in subsequent phases.

Witness Deepflare in action as it brings the intricate Rosetta model to life – a stunning showcase of Biotech and dataviz innovation.

How We Approached Communication with Deepflare

During the Deepflare project, natural questions arose from the client. Most discussions focused on:

UI Simplicity – the goal was to make the interface highly intuitive, without overwhelming researchers with extra elements.
Communication Between Frontend and Backend – ensuring smooth data flows and reliable API operation in complex scenarios.
API Design – it had to handle complex scenarios, including 3D synchronization, sequences, and region labeling.

Our Approach: Instead of long documents and theoretical plans, we worked iteratively. We created quick prototypes, conducted joint reviews, and made real-time adjustments. This allowed the client to see the application in action immediately, and we adapted it step by step to the user’s needs.

Summary – Positive Transformation & Project Successes

Thanks to this method, a product was created that now allows researchers to work independently, without coding. AI predictions are seamlessly integrated with interactive 3D visualization, and the entire system is modular and scalable, ready to accommodate additional models and levels of automation.

“The most challenging part was simplifying complex operations into a simple interface. We focused on clear paths and minimalism so scientists could concentrate on research, not managing the application.” – Łukasz Musiał, Software Developer

Today, Deepflare is a fully functional platform allowing researchers to:

Define input data and analyze AI results without technical support
Visualize protein structures in 3D and track changes in real time
Collaborate with various AI models in a single, unified environment

The product not only meets the initial objectives but also provides a solid foundation for further development – introducing new models, automating processes, and simplifying researchers’ work in future stages.