Editor’s note: Today’s post is by Mike Francomb, Senior Vice President of Technology, RecruitMilitary and a U.S. Army Veteran. RecruitMilitary is a wholly owned subsidiary of Bradley-Morris, Inc. (BMI), the largest military-focused recruiting company in the United States. RecruitMilitary uses Google Cloud Talent Solution to power its job search experience and connect more organizations with veteran talent.

For seven years, I served in the U.S. Army as a Field Artillery Officer, Military Occupation Code 13A. My time in service included a deployment to Operation Desert Shield / Desert Storm with the 24th Infantry Division out of Fort Stewart, GA, and a variety of front line artillery leadership roles, serving as a logistics officer for my unit and as an instructor teaching new officers how to be professional artillerymen. My day-to-day entailed leading teams of highly trained soldiers and managing logistics and materials to help those soldiers perform at a high level in stressful, fast-paced environments. It was my job to ensure we were ready to handle any circumstance. 

The hardest part about transitioning out of the Army in May 1996 as a highly trained artillery veteran was the fact that, though I felt prepared for any challenge ahead, I wasn’t sure I was making the right choice. I made a common mistake of transitioning veterans, I jumped right into an entrepreneurial venture. Looking back, I wish I’d had access to resources that displayed career options that would help translate my skills for the corporate world, it would have helped me be better prepared and know what my options were. I wasn’t ready to jump from the Army into running a business, and it was a long two years. 

Though my first job out of the Army was challenging, it taught me that I loved the start-up environment, and I joined RecruitMilitary in October 1998 when it was five months old. For the past 21 years, I have been fortunate enough to play an important role in helping RecruitMilitary grow to what it is today, the industry leader in connecting military veterans with organizations.  

RecruitMilitary connects organizations with veteran talent through over 30 products and services, all of which are fueled by our job board. Our job board, with over 1,400,000 members, is core to our business. In fact, if we don’t have an active and growing job board population, we don’t have the supply of veteran talent we need to deliver to our clients across our suite of services. 

With veteran unemployment at a 50-year low, it became increasingly challenging for RecruitMilitary to grow our veteran job seeker database and keep those veterans actively applying to client jobs. Being a data-driven company, we saw our existing search functionality was no longer producing the desired results for clients and began to receive client feedback about decreased candidate activity.

It was clear to us that we needed to begin adopting machine learning and more advanced search capabilities into our products and operations. The HR Tech space is shifting that way fast, and we want to be at the forefront. As we researched paths to take and learned of Google’s operating philosophy leading with AI, and that they were developing a tool for veteran job search, it made a lot of sense to go with a leader.

When Grow with Google announced its commitment to support veterans, we learned that we could add their military occupation code (MOS) translation feature to our job board through Cloud Talent Solution. This feature lets transitioning service members enter their military occupation codes (MOS, AFSC, NEC, or rating) directly into our search bar to see relevant civilian jobs available at client companies. We’re also using Cloud Talent Solution’s remote work functionality to provide an improved job search experience that allows our customers to make remote work opportunities in the U.S. more discoverable on their career sites. We’re excited about this feature, as it enhances our ability to deliver meaningful jobs to important members of our military community, military spouses, and veterans with limited mobility. 

The results of Cloud Talent Solution compared to our previous search are tremendous. Our job seekers are getting a much better experience, and to us that means more veterans are connected to jobs with our clients. We know this because our number of daily job applications has grown by 78 percent. And knowing that we now have a tool that’s going to learn and get better as more of our job seekers use it means that we will reap benefit for work done over time, and so will our clients and veterans who use our job board. That’s tremendous ROI to receive for a lean development staff. 

These are just a few of the types of tools I wish I’d had access to when I was considering my transition in 1996. With the help of technology and resources, like those from RecruitMilitary and Grow with Google, people in the military community, including veterans like myself, can prepare for and build meaningful careers.

Investing in Artificial Intelligence (AI) can bring a competitive advantage to your organization. If you’re in charge of an AI or Data Science team, you’ll want to measure and maximize the value that you’re providing. Here is some advice from our years of experience in the field. 

A checklist to embark on a project: 

As you embark on projects we’ve found it’s good to have the following areas covered: 

• Have a customer. It’s important to have a customer for your work, and that they  agree with what you’re trying to achieve. Be sure to know what value you’re delivering to them. 
• Have a business case.  This will rely on estimates and assumptions, and may take no more than a few minute’s work.  You should revise this, but always know what justifies your team’s effort, and what you (and your customer) expect to get in return. 
• Know what process you will change or create. You’ll want to put your work in production, so you have to be clear about what business operations are changing or created around your work and who needs to be involved to make it happen
• Have a measurement plan. You’ll want to show that ongoing work is impacting some relevant business indicator. Measure and show incremental value. The goal of these measurements is to establish what has changed because of your project that would otherwise not have changed. Be sure to account for other factors like seasonality or other business changes that may affect your measurements.
• Use all the above to get your organization’s support for your team and your work. 

What measures to use?

As you start the work, what measures and indicators can you use to show that your team’s work is useful for your organization?

How many decisions you make. A major function of ML is to automate and optimize decisions: which product to recommend, which route to follow, etc. Use logs to track how many decisions your systems are making. 

Changes to revenue or costs. Better and quicker decisions often lead to increased revenue or savings. If possible, measure it directly, otherwise estimate it (for example fuel costs saved from less distance traveled, or increased purchases from personalized offers). 

As an example, the Illinois Department of Employment Security is using Contact Center AI to rapidly deploy virtual agents to help more than 1 million citizens file unemployment claims. To measure success the team tracked the two outcomes:  (1) the number of web inquiries and voice calls they were able to handle, and (2) the overall cost of the call center after the implementation. Post implementation, they were able to observe more than 140,000 phone and web inquiries per day and over 40,000 after-hours calls per night. They  also anticipate an estimated annual cost savings of $100M based on an initial analysis of IDES’s virtual agent data (see more in the link to case study).

Implementation costs. The other side of increased revenue or savings, is to put your achievements in the context of how much they cost. Show the technology costs that your team incurs and, ideally, how you can deliver more value, more efficiently. 

How much time was saved.  If the team built a routing system then it saved travel time, if it built an email classifier then it saved reading time, etc. Quantify how many hours were given back to the organization thanks to the efficiency of your system. 

In the medical field, quicker diagnostics matter. Johns Hopkins University’s Brain Injury Outcomes (BIOS) Division has focused on studying brain hemorrhage aiming to improve medical outcomes. The team identified the time to insights as a key metric in measuring business success. They experimented with a range of cloud computing solutions like Dataflow, Cloud Healthcare API, Compute Engine, and AI Platform for distributed training to accelerate iterations. As a result, in their recent work they were able to accelerate insights from scans from approximately 500 patients from 2,500 hours to 90 minutes.

How many applications your team supports. Some of your organization’s operations don’t use ML (say reconciling financial ledgers) but others do. Know how many parts of your organization benefit from the optimization and automation your team builds.

User experience. You may be able to measure your customer’s experience: fewer complaints, better reviews, reduced latency, more interactions, etc. This is valid both for internal and external stakeholders. At Google we measure usage and regularly ask for feedback on any internal system or process.

One of our customers, The City of Memphis, is using VisionAI and ML to tackle a common but very challenging issue: identifying and addressing potholes.  The implementation team identified the percentage increase of potholes identified as one of the key metrics along with accuracy and cost savings. The solution captures video footage from it’s public vehicles and leverages Google Cloud capabilities like Compute Engine, AI Platform, and BigQuery to automate the review of videos.  The project increased  pothole detection by 75% with over 90% accuracy. By measuring and demonstrating these outcomes, the team proved the viability of a cost-effective, cloud-based machine learning model and is looking into new applications of AI and ML that will further improve city services and help it build a better future for its 652,000 residents. 

Acknowledgements

Filipe and Payam would like to thank our colleague and co-author Mona Mona (AI/ML Customer Engineer, Healthcare and lifesciences) who contributed equally to the writing.

Today, to accelerate research in the bio-pharma space, from the creation of treatments for diseases to the production of new synthetic biomaterials, we are announcing a new Vertex AI solution that demonstrates how to use Vertex AI Pipelines to run DeepMind’s AlphaFold protein structure predictions at scale.

Once a protein’s structure is determined and its role within the cell is understood, scientists can develop drugs that can modulate the protein function based on its role in the cell. DeepMind, an AI research organization within Alphabet, created the AlphaFold system to advance this area of research by helping data scientists and other researchers to accurately predict protein geometries at scale.

In 2020, in the Critical Assessment of Techniques for Protein Structure Prediction (CASP14) experiment, DeepMind presented a version of AlphaFold that predicted protein structures so accurately, experts declared the “protein-folding problem” solved. The next year, DeepMind open sourced the AlphaFold 2.0 system. Soon after, Google Cloud released a solution that integrated AlphaFold with Vertex AI Workbench to facilitate interactive experimentation. This made it easier for many data scientists to efficiently work with AlphaFold, and today’s announcement builds on that foundation.

Last week, AlphaFold took another significant step forward when DeepMind, in partnership with the European Bioinformatics Institute (EMBL-EBI), released predicted structures for nearly all cataloged proteins known to science. This release expands the AlphaFold database from nearly 1 million structures to over 200 million structures—and potentially increases our understanding of biology to a profound degree. Between this continued growth in the AlphaFold database and the efficiency of Vertex AI, we look forward to the discoveries researchers around the world will make.

In this article, we’ll explain how you can start experimenting with this solution, and we’ll also survey its benefits, which include offering lower costs through optimized selection of hardware, reproducibility through experiment tracking, lineage and metadata management, and faster run time through parallelization.

Background for running AlphaFold on Vertex AI

Generating a protein structure prediction is a computationally intensive task. It requires significant CPU and ML accelerator resources and can take hours or even days to compute. Running inference workflows at scale can be challenging—these challenges include optimizing inference elapsed time, optimizing hardware resource utilization, and managing experiments.Our new Vertex AI solution is meant to address these challenges.

To better understand how the solution addresses these challenges, let’s review the AlphaFold inference workflow:

1. Feature preprocessing. You use the input protein sequence (in the FASTA format) to search through genetic sequences across organisms and protein template databases using common open source tools. These tools include JackHMMER with MGnify and UniRef90, HHBlits with Uniclust30 and BFD, and HHSearch with PDB70. The outputs of the search (which consist of multiple sequence alignments (MSAs) and structural templates) and the input sequences are processed as inputs to an inference model. You can run the feature preprocessing steps only on a CPU platform. If you’re using full-size databases, the process can take a few hours to complete.
2. Model inference. The AlphaFold structure prediction system includes a set of pretrained models, including models for predicting monomer structures, models for predicting multimer structures, and models that have been fine-tuned for CASP. At inference time, you independently run the five models of a given type (such as monomer models) on the same set of inputs. By default, one prediction is generated per model when folding monomer models, and five predictions are generated per model when folding multimers. This step of the inference workflow is computationally very intensive and requires GPU or TPU acceleration.
3. (Optional) Structure relaxation. In order to resolve any structural violations and clashes that are in the structure returned by the inference models, you can perform a structure relaxation step. In the AlphaFold system, you use the OpenMM molecular mechanics simulation package to perform a restrained energy minimization procedure. Relaxation is also very computationally intensive, and although you can run the step on a CPU-only platform, you can also accelerate the process by using GPUs.

The Vertex AI solution

The AlphaFold batch inference with the Vertex AI solution lets you efficiently run AlphaFold inference at scale by focusing on the following optimizations:

• Optimizing inference workflow by parallelizing independent steps.
• Optimizing hardware utilization (and as a result, costs) by running each step on the optimal hardware platform. As part of this optimization, the solution automatically provisions and deprovisions the compute resources required for a step.
• Describing a robust and flexible experiment tracking approach that simplifies the process of running and analyzing hundreds of concurrent inference workflows.

The following diagram shows the architecture of the solution.

The solution encompasses the following:

A strategy for managing genetic databases. The solution includes high-performance, fully managed file storage. In this solution, Cloud Filestore is used to manage multiple versions of the databases and to provide high throughput and low-latency access.

An orchestrator to parallelize, orchestrate, and efficiently run steps in the workflow. Predictions, relaxations, and some feature engineering can be parallelized. In this solution, Vertex AI Pipelines is used as the orchestrator and runtime execution engine for the workflow steps.

Optimized hardware platform selection for each step. The prediction and relaxation steps run on GPUs, and feature engineering runs on CPUs. The prediction and relaxation steps can use multi-GPU node configurations. This is especially important for the prediction step because the memory usage is approximately quadratic with the number of residues. Therefore, predicting a large protein structure can exceed the memory of a single GPU device.

Metadata and artifact management. The solution includes management for running and analyzing experiments at scale. In this solution, Vertex AI Metadata is used to manage metadata and artifacts.

The basis of the solution is a set of reusable Vertex AI Pipelines components that encapsulate core steps in the AlphaFold inference workflow: feature preprocessing, prediction, and relaxation. In addition to those components, there are auxiliary components that break down the feature engineering step into tools, and helper components that aid in the organization and orchestration of the workflow.

The solution includes two sample pipelines: the universal pipeline and a monomer pipeline. The universal pipeline mirrors the settings and functionality of the inference script in the AlphaFold Github repository. It tracks elapsed time and optimizes compute resources utilization. The monomer pipeline further optimizes the workflow by making feature engineering more efficient. You can customize the pipeline by plugging in your own databases.

Next steps

To learn more and to try out this solution, check our GitHub repository, which contains the components and universal and monomer pipelines. The artifacts in the repository are designed so that you can customize them. In addition, you can integrate this solution into your upstream and downstream workflows for further analysis. To learn more about Vertex AI, visit our product page.

Acknowledgements

We would like to thank the following people for their collaboration: Shweta Maniar, Sampath Koppole, Mikhail Chrestkha, Jasper Wang, Alex Burdenko, Meera Lakhavani, Joan Kallogjeri, Dong Meng (NVIDIA), Mike Thomas (NVIDIA), and Jill Milton (NVIDIA).

Finally and most importantly, we would like to thank our Solution Manager Donna Schut for managing this solution from start to finish. This would not have been possible without Donna.