Unlocking Secure, Flexible AI Deployment Across Environments

Following the recent announcement of Revilico Inc.’s partnership with M9, we’re excited to share a significant leap forward in accelerating high-throughput screening by integrating their inference-at-source optimization framework. This upgrade not only delivers breakthrough performance across our platform but does so while keeping your most sensitive data secure and under control. The result is a new standard in drug discovery AI—where performance, privacy, and flexibility are in perfect unity.

One of the most transformative outcomes of this integration is the ability to run high-speed AI inference across multiple environments. Our platform now supports full deployment in the cloud, on-premise infrastructure, or even directly on a personal workstation. This multi-modality ensures that our clients—whether startups with agile workflows or enterprises with strict regulatory protocols—can operate wherever they are most comfortable. For many, the cloud remains the most scalable and efficient option, and with M9’s optimization layer, our SaaS performance has seen great improvements. But now, we also open the door to on-prem and hybrid models, empowering teams to run AI securely within their own environments without giving up speed or scale.

With M9’s ability to bring intelligence closer to the data, organizations gain full control over how and where their drug discovery workflows are executed. Whether protecting proprietary compound libraries, complying with regional data laws, or safeguarding internal omics datasets, our platform adapts to your needs. The same Revilico experience is now available across all deployment modes—cloud-native, on-prem, or local—without sacrificing model performance, usability, or throughput.

Smarter Inference, Scalable Performance, Less Hardware Overhead

This flexibility is matched by technical excellence. Traditionally, calculating essential activity metrics using docking or molecular dynamics across large compound libraries demanded significant time and hardware. With M9’s intelligent inference engine embedded in our platform, we’ve drastically improved compute efficiency, optimizing matrix operations, memory access, and prediction caching to dramatically reduce run times and hardware requirements. Instead of scaling GPUs endlessly, we focused on scaling smarter.

On standard GPU infrastructure (g4dn.xlarge), we observed a 13x speedup in prediction throughput. Scaling up to a g4dn.2xlarge instance, we reached rates exceeding 130 compounds per second—a performance ceiling that redefines expectations for AI in pharma. These gains are not theoretical—they’ve been rigorously benchmarked across large compound screens.

Benchmarking Results and Docking Comparisons

Computational Benchmarks:

• CPU (estimated): 4.34 compounds/sec

• Traditional GPU: 5.21 compounds/sec

• GPU + M9 (Cold Start): 40.65 compounds/sec

• GPU + M9 (Hot Start): 69.44 compounds/sec

• GPU + M9 (Hot, 2xlarge GPU): 139.60 compounds/sec

This chart illustrates a significant performance breakthrough in activity-based virtual screening. When benchmarked against traditional docking platforms, Revilico’s optimized inference engine on a single g4dn.2xlarge instance achieved a throughput of 139.6 compounds per second—more than 2x the speed of AutoDock GPU (63.3 cmp/sec) running on the Summit supercomputer, and vastly outperforming both Glide HTVS (5.6 cmp/sec) and AutoDock Vina on CPU (2.0 cmp/sec).

These results underscore how our platform, through software-based optimization and inference-at-source architecture, surpasses legacy docking workflows that typically rely on computationally expensive grid search or physics-heavy scoring functions. Instead of needing massive clusters or high-end HPC, Revilico delivers state-of-the-art screening speeds on widely available cloud hardware, making real-time molecular assessment feasible for any lab, anywhere.

Operational Impact and Vision for the Future

These results aren't just technical wins—they're operational game-changers. Workloads that once consumed hours now finish in minutes. Entire high-throughput campaigns can now be executed efficiently on compact, secure infrastructure—or scaled seamlessly in the cloud.

The strategic implications for R&D teams are vast. With faster inference, teams can iterate on lead compounds more frequently, explore analog series in real time, and accelerate SAR cycles. Ultra-large-scale screening becomes practical, enabling direct exploration of Revilico’s internally hosted library of 70 billion virtual molecules. And for organizations with constrained compute budgets, this performance shift creates new opportunities to compete and innovate using our optimized cloud environment, local infrastructure, or hybrid setups.

Crucially, this advancement doesn’t stop at molecular affinity prediction. The performance boost is being embedded throughout our pipeline—across virtual screening, docking, ADMET filtering, multi-objective lead optimization, and biologically aware modeling. In every module, we’re removing friction from the Design–Make–Test–Analyze (DMTA) cycle and bringing high-impact drug discovery tools directly to the hands of medicinal chemists, data scientists, and translational teams alike.

Looking ahead, our vision remains the same: to deliver AI-powered discovery tools that accelerate science, respect data boundaries, and adapt to the evolving needs of drug hunters everywhere. Whether your workflow is cloud-first, compliance-driven, or entirely local, Revilico is ready to meet you there—with uncompromising intelligence, privacy, security, and speed.

If your team is ready to explore what intelligent deployment and smart infrastructure can do for your next therapeutic campaign, we welcome the conversation.

Let’s build a faster, more secure, and more adaptive future for drug discovery—together.

🔗 Revilico 🔗 M9