At FullTilt: Real-Time Open-Set 3D Macromolecule Detection Directly from Tilted 2D Projections

FullTilt represents a significant shift in cryo-electron tomography (cryo-ET) workflows. Traditional macromolecule detection requires reconstructing a 3D tomogram first, which is computationally expensive and introduces VRAM bottlenecks for deep learning models. By operating directly on aligned 2D tilt-series, FullTilt circumvents the reconstruction overhead. The 'open-set' capability is its primary differentiator, allowing researchers to detect novel proteins without specific retraining—a massive pain point in structural biology. From a competitive standpoint, its defensibility (5) is currently rooted in domain expertise and the niche nature of cryo-ET data structures rather than community momentum (0 stars, 5 days old). It competes with established tools like crYOLO and DeepFinder, but offers a distinct architectural advantage for high-throughput pipelines. Frontier labs like OpenAI or Google (excluding DeepMind's specialized units) are unlikely to compete here as the data is too specialized and the market is purely academic/R&D. The primary risk is consolidation within the structural biology software ecosystem (e.g., integration into Scipion or RELION frameworks), which would turn this from a standalone tool into a plugin. Its displacement horizon is long because the specialized math required to map 2D projections to 3D coordinates in a low SNR environment provides a technical moat against generic computer vision approaches.