Zero-shot mutation effect prediction on protein stability and function using RoseTTAFold

Predicting the effects of mutations on protein function and stability is an outstanding challenge. Here, we assess the performance of a variant of RoseTTAFold jointly trained for sequence and structure recovery, RFjoint , for mutation effect prediction. Without any further training, we achieve comparable accuracy in predicting mutation effects for a diverse set of protein families using RFjoint to both another zero-shot model (MSA Transformer) and a model that requires specific training on a particular protein family for mutation effect prediction (DeepSequence). Thus, although the architecture of RFjoint was developed to address the protein design problem of scaffolding functional motifs, RFjoint acquired an understanding of the mutational landscapes of proteins during model training that is equivalent to that of recently developed large protein language models. The ability to simultaneously reason over protein structure and sequence could enable even more precise mutation effect predictions following supervised training on the task. These results suggest that RFjoint has a quite broad understanding of protein sequence-structure landscapes, and can be viewed as a joint model for protein sequence and structure which could be broadly useful for protein modeling.

20th-Century hurricanes leave long-lasting legacies on tropical forest height and the abundance of a dominant wind-resistant palm

Projected increases in hurricane intensity under a warming climate will have profound effects on many forest ecosystems. One key challenge is to disentangle the effects of wind damage from the myriad factors that influence forest structure and species distributions over large spatial scales. Here, we employ a novel machine learning framework with high-resolution aerial photos, and LiDAR collected over 115 km2 of El Yunque National Forest in Puerto Rico to examine the effects of topographic exposure to two hurricanes, Hugo (1989) and Georges (1998), and several landscape-scale environmental factors on the current forest height and abundance of a dominant, wind-resistant species, the palm Prestoea acuminata var. montana. Model predictions show that the average density of the palm was 32% greater while the canopy height was 20% shorter in forests exposed to the two storms relative to unexposed areas. Our results demonstrate that hurricanes have lasting effects on forest canopy height and composition, suggesting the expected increase in hurricane severity with a warming climate will alter coastal forests in the North Atlantic.

Increasing taxonomic diversity and spatial resolution clarifies opportunities for protecting US imperiled species

Continental- and regional-scale assessments of gaps in protected area networks typically use relatively coarse range maps for well documented species groups, creating uncertainty about the fate of unexamined biodiversity and providing insufficient guidance for land managers. By building habitat suitability models for a taxonomically diverse group of 2216 imperiled plants and animals, we revealed comprehensive and detailed protection opportunities in the conterminous United States. Summing protection-weighted range-size rarity (PWRSR, the product of the percent of modeled habitat outside of protected areas and the inverse of modeled habitat extent) uncovered novel patterns of biodiversity importance. Concentrations of unprotected imperiled species in places such as the northern Sierra Nevada, central and northern Arizona, the Rocky Mountains of Utah and Colorado, southeastern Texas, southwestern Arkansas, and Florida's Lake Wales Ridge have rarely if ever been featured in continental- and regional-scale analyses. Inclusion of diverse taxa (vertebrates, freshwater mussels, crayfishes, bumble bees, butterflies, skippers, and vascular plants) partially drove these new patterns. When analyses were restricted to groups typically included in previous studies (birds, mammals, and amphibians), up to 53% of imperiled species in other groups were left out. The finer resolution of modeled inputs (990 m) also resulted in a more geographically dispersed pattern. For example, 90% of the human population of the conterminous United States lives within 50 km of modeled habitat for one or more species with high PWRSR scores. Over one-half of the habitat for 818 species occurs within federally lands managed for biodiversity protection; an additional 360 species have over one-half of their modeled habitat on federal multiple use land. Freshwater animals occur in places with poorer landscape condition but with less exposure to climate change than other groups, suggesting that habitat restoration is an important conservation strategy for these species. The results provide fine-scale, taxonomically diverse inputs for local and regional priority-setting and show that although protection efforts are still widely needed on private lands, notable gains can be achieved by increasing protection status on selected federal lands.