Stealth Series-B Startup
Trailblazing Biotech startup accelerated DMTA cycles at reduced cloud expense
Fovus supercharged our computational drug discovery pipelines. Fovus accelerated our time-to-insight in DMTA cycles by 96x, at one-fifth of the overall cloud cost.
Director of Computational Chemistry
Optimal
HPC
strategy
Massive
task-level
parallelism
Spot
instance
auto-leveraged
4x
faster
DMTA cycles
5.5x
lower
cloud cost
Optimal
HPC strategy
Massive
task-level parallelism
Spot
instance auto-leveraged
4x
faster DMTA cycles
5.5x
lower cloud cost
| Self-managed AWS Time | Self-managed AWS Cost | Fovus Time | Fovus Cost | Time Reduction | Cost Saving | |
|---|---|---|---|---|---|---|
| HTVS (Docking) | 45 hrs | $3,250 | 3.5 hrs | $650 | 13x | 5x |
| AI-augmented HTVS (Training) | 24 hrs (local) | – | – | – | – | – |
| AI-augmented HTVS (Prediction) | 48 hrs | $800 | 0.5 hrs | $100 | 96x | 8x |
| Total (5 DMTA cycles) | 24 days | $20,250 | 6 days | $3,750 | 4x | 5.5x |
About Company
A stealth series-B biotechnology startup focused on developing innovative and efficacious therapies to make them accessible to every patient.
Industry
Biotechnology.
HPC Workloads
High-throughput virtual screening (HTVS), molecular dynamics simulation, molecular docking simulation, AI-augmented HTVS, etc.
Pains
- Managing cloud logistics for computational drug discovery is time-consuming and distracting, causing 10-20% innovation downtime.
- Cloud offerings are too diverse and heterogeneous, making it impossible to find the optimal HPC strategies for various workloads.
- A suitable HPC strategy for workload A could be bad for workload B. Without proper HPC strategies, the team experiences high cloud HPC costs, especially for GPU workloads.
- Scaling up cluster size is difficult due to a lack of cloud expertise, especially for GPU workloads. This limits the computation scale and leads to unbearably long time-to-insight for large-scale screening, slowing down DMTA cycles.
Solution
Serverless HPC experience, ease-of-everything
- Fovus uses AI to automatically orchestrate all the cloud logistics and optimize the HPC strategies so the team can say goodbye to all the cloud integration and management hassles.
- Fovus offers a simple, out-of-the-box web UI and CLI. Tapping into supercomputing power is as simple as a single command or a few clicks. Users only deploy, and Fovus takes care of everything else.
- Easy to integrate, deploy, scale, optimize, and sustain.
- Fovus also offers Python API for the team to integrate serverless HPC capabilities into their in-house software applications.
Free automated benchmarking
- Fovus auto-benchmarked each type of the team’s computational chemistry workload for free to explore their HPC strategy space effectively.
- This reveals how the choice of CPUs/GPUs, hyperthreading, memory, and storage configurations impact the performance and cost of each computational chemistry workload.
- Such benchmarking data provides the foundation for optimizing the HPC strategy tailored to each computational chemistry workload.
AI-optimized HPC strategy
- Based on the custom benchmarking data, Fovus auto-determines the optimal HPC strategy to minimize the total runtime and cloud cost for each computational chemistry workload.
- This ensures the time-cost optimality of computational drug discovery with zero user effort needed.
Multi-cloud-zone auto-scaling
- Fovus auto-scales the cluster size and the tasks running in parallel across multiple cloud availability zones according to their availability dynamics to leverage abundant cloud resources best to accelerate computational drug discovery.
- This massively parallelized the computation for the team’s large-scale HTVS tasks and reduced the overall time-to-insight of a critical step in DMTA cycles by nearly 100x.
Continuously improve HPC strategy
- Fovus auto-updates the custom benchmarking data and continuously improves the HPC strategies for each computational chemistry workload as cloud infrastructure constantly upgrades.
- Take the HTVS workload as an example.
- The initial HPC strategy optimization improved performance by 1.6x and lowered cloud cost by 2x at the task level compared to the prior user HPC strategy by leveraging the state-of-the-art ARM CPUs.
- 6 months later, Fovus auto-upgraded the HPC strategy as new cloud infrastructure was rolled out with the next-generation x86 CPU architecture. This brought another 1.5x faster performance and 1.3x lower cost at the task level.
- Such performance and cost continued to improve as hardware technology quickly evolved.
Auto-leverage spot instance
- For each computational chemistry workload, Fovus intelligently analyzes the expected savings for each applicable spot strategy according to the availability and pricing dynamics of spot instances, their frequency of interruption, and workload runtime predictions.
- Fovus automatically leverages spot instances by applying the best spot strategy to further reduce the cloud cost by another 2-3x, making computational drug discovery highly affordable at scale.
- Fovus automatically re-queues any disrupted tasks due to reclaiming spot instances for re-execution to assure the integrity of discovery results with failover capabilities.
Results
- Enjoy the freedom from cloud integration and management hassles.
- 15% more work productivity. More valuable time is spent on science and discovery.
- Significantly faster and cheaper computational drug discovery (see table below)
- For docking-based HTVS, time-to-insight was reduced from 2 days to 3.5 hours—13x faster. Cloud cost was reduced from $3,250 to $650 per launch—5x lower expense.
- For AI-augmented HTVS, time-to-insight was reduced from 2 days to 30 minutes—96x faster. Cloud cost was reduced from $800 to $100 per launch—8x lower expense.
- 8000-node cluster launched across multiple availability zones in less than 20 minutes—all tasks ran in parallel.
- Overall, 4x faster DMTA cycles were achieved at 5.5x less cloud spending.
- Ability to consistently leverage state-of-the-art hardware technology to supercharge computational drug discovery and stay ahead of the game.
Other case studies
Success Story
World’s 2nd largest heavy equipment manufacturer accelerated digital innovation
Success Story
Leading discovery service provider accelerated DMTA cycles and project delivery at reduced costs
