About This Project

Atmospheric methane, contributing to 30% of global warming, remains challenging to mitigate due to its low concentration. We propose developing enhanced soluble methane monooxygenase variants optimized for atmospheric methane conversion through AI protein optimization. By combining AI modeling with high-throughput screening, we aim to achieve a 100-fold increase in methanol production efficiency at atmospheric methane levels, enabling economically viable capture from dilute sources.

Ask the Scientists

Motivating Factor

Methane (CH₄) accounts for 30% of current global warming[1], yet 75% of emissions occur at concentrations too dilute (<1,000 ppm) for cost-effective mitigation[2]. Deployment of methane monooxygenases (MMO) in either bioreactors or in engineered crops could oxidize CH₄, but existing MMO lack the catalytic efficiency for practical deployment: expression of mmoXYBZDC[3] achieves methane conversion, but only at high methane concentrations. Therefore, improving MMO through protein engineering is critical for large-scale CH₄ oxidation. Traditional protein engineering relies on methods that sample a fraction of sequence space, making optimization difficult. Our AI platform screens >10⁷ variants in silico per cycle, consistently achieving 100-500× activity gains while maintaining protein stability. This approach could reduce reactor costs for oxidizing 500 ppm CH₄ to economically viable levels ($100/t CO₂e target[4]) in just 6 cycles, versus years of traditional evolution.

Specific Bottleneck

Soluble MMO (sMMO) fail due to both low expression/activity. The heterologously expressed mmoXYBZDC [3] is optimized for industrial CH₄ removal (550,000 ppm) and requires multiple improvements:

1. Unsuitability for trace CH₄: At 550,000 ppm, CH₄ saturates the enzyme, masking its poor activity at 500 ppm. For dilute CH₄, activity must increase by ≥100× to achieve economic viability.

2. Tradeoffs in evolution: Lowering kₘ to <10 nM (required for 500 ppm[4]) risks destabilizing the enzyme or reducing Vₘₐₓ. Conventional mutagenesis cannot test enough variants to resolve this tradeoff.

3. Lack of low-CH₄ data: mmoXYBZDC's kinetics (kₘ, Vₘₐₓ) at <1,000 ppm CH₄ are uncharacterized. This data gap critically hampers rational engineering efforts and prevents accurate technoeconomic assessment of mmoXYBZDC for climate applications.

4. Poor heterologous expression: mmoXYBZDC requires stabilization, and may not be suitable for long term expression in either E. coli or other potential hosts.

Actionable Goals

We will deploy our EVOLVEproV2 AI model—a protein evolution platform that simultaneously optimizes multiple properties through multi-objective learning across diverse experimental data. By screening >10⁷ variants in silico per cycle, EVOLVEproV2 achieves 100–500× gains across multiple features concurrently. Goals include:

1. 100× higher specific affinity: Use EVOLVEproV2 to co-evolve mmoXYBZDC's a˚S ≥100 μM⁻¹s⁻¹ (vs. ~1 μM⁻¹s⁻¹ in wild-type sMMO [16]) via optimization of kₘ (<10 nM) and Vₘₐₓ (>1 μmol/min/mg). This enables economically viable oxidation of 500 ppm CH₄ at ≤$100/t CO₂e [4].

2. Enhanced stability: Leverage EVOLVEpro to maintain robust mmoXYBZDC and GroESL scaffold stability [3] while improving catalytic efficiency.

3. Precise substrate specificity: Engineer ≥95% reduction in unwanted methanol oxidation.

Our multi-objective AI approach will accomplish these interlinked goals in 6 cycles versus years of traditional evolution.