Simulating Neutron Protein Crystallography Experiments: Applications to the Development of the NMX Instrument at ESS

This paper presents Monte Carlo simulations using McStas to optimize the upcoming NMX instrument at the European Spallation Source by implementing ray-splitting techniques and a new sampling method to improve event formation and assess environmental scattering effects for neutron protein crystallography.

The Gist

Imagine you are an architect designing the world's most powerful flashlight, but you haven't built the building yet, and you don't have any light bulbs to test it with. How do you know if your design will work? You build a perfect, hyper-realistic virtual simulation on a supercomputer.

That is exactly what this paper is about. The authors are simulating a neutron experiment for a massive new machine called NMX at the European Spallation Source (ESS), all without firing a single real neutron.

Here is the breakdown of their work using everyday analogies:

1. The Goal: Seeing the Invisible

Scientists want to study proteins (the tiny machines inside our bodies) to understand how they work. To do this, they need to see where the hydrogen atoms are located.

• The Problem: X-rays (like in medical CT scans) are great at seeing heavy atoms like carbon, but they "glance off" hydrogen atoms, making them invisible.
• The Solution: Neutrons are like a different kind of flashlight that interacts perfectly with hydrogen. However, neutrons are hard to get; they are rare and weak compared to X-rays.
• The Challenge: To get a good picture, you need huge crystals and a lot of time. The new ESS machine promises to be so powerful it can use smaller crystals, but before they build it, they need to know exactly how to set it up.

2. The Tool: The "Virtual Ray-Tracer"

The team used a software called McStas. Think of this as a video game engine for physics.

• Instead of shooting real neutrons, the computer shoots millions of virtual neutron rays.
• These rays travel through a virtual model of the machine, bouncing off mirrors, passing through choppers (gates that control the beam), and hitting a virtual protein crystal.
• The computer calculates the odds of every single ray hitting a specific spot on a virtual detector.

3. The Big Hurdle: The "Needle in a Haystack" Problem

Here is the tricky part: The probability of a neutron hitting a protein crystal and bouncing off in a useful way is incredibly low. It's like trying to hit a specific pixel on a wall by throwing a dart from a mile away.

• The Old Way: To get enough "hits" (data) to see a clear picture, you would have to simulate trillions of darts. This would take a supercomputer years to calculate.
• The New Trick (SPLIT): The authors found a clever shortcut. Imagine you throw one dart, and it hits the wall. Instead of just recording that one hit, the computer says, "Okay, let's copy and paste that successful dart 10,000 times."
  • This is called the SPLIT method.
  • It allows them to simulate the results of trillions of darts by only actually calculating the path of a few million. It's like having a photocopier for your data, saving them massive amounts of time and computer power.

4. Turning "Probabilities" into "Real Data"

The computer doesn't give you a picture; it gives you a list of probabilities (e.g., "There is a 0.0000001% chance a neutron hit this pixel").

• The Problem: You can't analyze a list of probabilities the same way you analyze a photo. You need actual "counts" (like a real detector would see).
• The Solution: They used a new mathematical trick called Weighted Reservoir Sampling.
  • Analogy: Imagine you have a giant river of water (the probabilities) flowing past you. You can't drink the whole river, but you need a cup of water that represents the whole river perfectly.
  • This new method lets them scoop up a "cup" of data that statistically looks exactly like the real thing, without needing to store the entire river in their memory. This allows them to turn the simulation into a file that looks exactly like data from a real experiment.

5. Testing the "Real World" Messiness

In a perfect world, neutrons only hit the crystal. In reality, they bounce off air, the beamstop (a shield to catch the main beam), and the walls.

• The team simulated these messy conditions. They found that air is actually a big source of "noise" (static on a radio), scattering neutrons and creating a fuzzy background.
• By simulating this, they learned exactly how to position the detectors and shields to minimize this noise before the real machine is even built.

6. The Result: A Blueprint for Success

They took their simulated data and ran it through the same software scientists use to analyze real experiments (called DIALS).

• The Verdict: The software successfully found the protein structure in the fake data.
• Why it matters: This proves their simulation is accurate. Now, when the real NMX machine opens in 2027, they won't be guessing how to set it up. They will have a "flight simulator" that tells them exactly where to put the detectors, how long to run the experiment, and how to clean up the data.

Summary

This paper is about building a digital twin of a giant scientific machine. By using clever math tricks (SPLIT and Reservoir Sampling), the team turned a computer simulation into a realistic "fake experiment." This allows them to perfect the design of the world's most powerful neutron microscope before a single bolt is tightened, ensuring that when it finally turns on, it will immediately start revealing the secrets of life at the atomic level.

Technical Summary

1. Problem Statement

Neutron macromolecular crystallography (n-MX) is a powerful technique for determining precise hydrogen atom positions in biological macromolecules, complementing X-ray and electron microscopy. However, it faces significant experimental challenges:

• Low Flux: Neutron sources have significantly lower flux than X-ray sources, requiring large, isotopically deuterated crystals (often >1 mm³) to obtain usable data.
• Instrument Complexity: The upcoming NMX instrument at the European Spallation Source (ESS) utilizes a unique variable-geometry detector system and time-of-flight (TOF) Laue techniques to maximize data collection efficiency.
• Lack of Real Data: As NMX is currently under construction, there is no real experimental data available to validate its design, optimize data collection strategies, or test data processing pipelines.
• Computational Barriers: Simulating n-MX experiments is computationally expensive due to the low probability of neutron scattering events and the need to sample vast reciprocal spaces (often >10⁶ reflections). Standard Monte Carlo methods struggle with the memory and processing power required to generate statistically significant datasets for complex instruments.

2. Methodology

The authors developed a comprehensive simulation workflow using the McStas software package (a Monte Carlo ray-tracing tool for neutrons) to model the NMX instrument and protein crystal diffraction.

• Instrument Modeling:
  • Source: Modeled the ESS "butterfly" moderator and the long pulse length (2.86 ms), which necessitates a long sample-to-source distance (157 m) for TOF resolution.
  • Optics: Included neutron guides, choppers (defining the 1.8–3.95 Å wavelength range), and collimation systems.
  • Scattering Components: Modeled air scattering and a realistic beamstop (a 6LiF powder disc in an Al casing) using the Union component system to handle complex geometries and multiple scattering.
  • Detectors: Simulated three mobile, wide-area GEM detector panels capable of 11 different configurations, covering the top hemisphere around the sample.
• Crystal Simulation:
  • Used Perdeuterated Pyrococcus furiosus rubredoxin (PDB 4AR4) as the test subject.
  • Generated a complete list of Bragg reflections (Miller indices and structure factors) using Gemmi and Phenix, ensuring Friedel pairs were included to simulate a full sphere of reciprocal space.
  • Implemented the SPLIT instruction in McStas: When a neutron ray hits the crystal, it is "split" into thousands of copies (e.g., SPLIT = 11,000). This drastically increases the number of recorded scattering events without re-calculating the trajectory from the source to the crystal, improving computational efficiency by orders of magnitude.
• Data Sampling (The "Event" Problem):
  • McStas outputs probabilities, not discrete events. To convert these into realistic datasets, the authors employed a single-pass weighted reservoir sampling algorithm (using the StreamSampling.jl library).
  • This method allows for the generation of a fixed-size dataset (reservoir) from a stream of probabilities without loading the entire dataset into memory, overcoming the terabyte-scale memory limitations of traditional multi-pass sampling.
• Data Processing:
  • Simulated data was converted to NeXus TOFRaw format.
  • Binned and histogrammed using the ESSNMX software (based on scipp).
  • Processed using DIALS (v3.27.0) for spot finding, indexing, and integration (using TOF-specific routines like dials.tof_integrate).

3. Key Contributions

• First-Principles n-MX Simulation: Demonstrated the feasibility of simulating a complete TOF-Laue n-MX experiment for a next-generation instrument (NMX) before it is built.
• Efficiency Optimization: Validated the use of the SPLIT instruction for crystal scattering, showing it can increase recorded event counts by ~10,000-fold with minimal computational overhead.
• Novel Sampling Algorithm: Introduced a single-pass weighted reservoir sampling method to convert Monte Carlo probabilities into discrete event data, solving the memory bottleneck associated with simulating large reciprocal spaces.
• Pipeline Integration: Established a full end-to-end pipeline from McStas simulation to DIALS processing, proving that simulated data can be treated identically to real experimental data.
• Background Analysis: Quantified the impact of non-crystal scattering (air, beamstop) on data quality, identifying air scattering as the dominant source of background noise, particularly for detectors placed far from the sample.

4. Results

• Event Generation: Using a SPLIT factor of 11,000, the simulation recorded ~2.3 × 10¹⁰ events from 5 × 10¹¹ source simulations, compared to only ~6.8 × 10⁵ events without SPLIT.
• Data Quality:
  • Simulated datasets processed with DIALS yielded high indexing rates (up to 98.7%) and resolution limits ($d_{min}$) down to 1.20 Å with sufficient sampling.
  • The signal-to-noise ratio ($I/\sigma(I)$) improved significantly with larger reservoir sizes, reaching values >100 for optimal sampling.
• Background Effects:
  • Simulations with a realistic beamstop and air scattering showed that air scattering is the primary contributor to background noise, creating a diffuse scattering cone around $d \approx 1.5$ Å.
  • The beamstop contribution was found to be relatively small.
  • Detector configurations placed further from the sample (e.g., Configurations 9 and 10) suffered significantly from exogenous scatter, dominating the signal.
• Performance: A complex simulation (5 × 10¹¹ rays) ran in ~38 minutes on 1280 CPUs (LUMI supercomputer), demonstrating the scalability of the approach.

5. Significance

This work provides a critical "digital twin" for the NMX instrument at ESS. By validating the simulation workflow against standard crystallographic software (DIALS), the authors have established a robust framework for:

1. Instrument Commissioning: Testing detector geometries and alignment strategies before the instrument is operational.
2. Data Collection Strategy: Optimizing sample-to-detector distances and exposure times to minimize background noise from air scattering.
3. Methodology Development: Proving that high-fidelity Monte Carlo simulations can replace early-stage experimental trials, saving time and resources in the development of neutron scattering facilities.
4. Error Modeling: Laying the groundwork for future studies on experimental errors (e.g., jitter, noise) in n-MX, which are difficult to model empirically.

The study confirms that with advanced sampling techniques and high-performance computing, it is possible to "perform a neutron experiment without neutrons," providing invaluable insights for the future of neutron macromolecular crystallography.