Multistate mechanisms underlie many of the complex functions observed in natural proteins. The ability to rationally design multistate proteins would have transformative implications for many areas of biotechnology, yet lies beyond the capabilities of existing deep learning frameworks for protein design. To address this gap, we introduce SwitchCraft, a versatile and programmatic framework for designing state-switching proteins based on backpropagation through compositional design constraints parameterized by structure prediction models. In silico evaluations demonstrate success on a wide range of state-switching functional primitives, from allosteric regulation of motifs to discrimination of bound ligand identities. Using these primitives, we demonstrate an in silico strategy for de novo design of fluorescent biosensors to arbitrary small molecule analytes. These results position SwitchCraft at the inception of a powerful paradigm for higher-order functional protein design. Code is available at https://github.com/bjing2016/switchcraft.

From the introduction: Introduction 2 Background Generative modeling for protein design. Multistate protein design. 3 Method 3.1 Design specification Motif loss. Binding loss. Conformational change loss. Contact loss. 3.2 Design optimization 4 Experiments 4.1 Positive and negative allostery 4.2 Motif switching 4.3 Ligand modification 4.4 Induced binding 4.5 Ligand discrimination 4.6 Workflow for biosensor design 5 Conclusion References A Additional Method Details A.1 Design optimization A.2 Evaluation criteria. A.3 Motif specifications A.4 Biosensor design A.4.1 Additional Background A.4.2 Design Clarifications B Additional Results B.1 Positive and negative allostery B.2 Motif switching B.3 Ligand modification B.4 Induced binding B.5 Ligand discrimination B.6 Controls & Ablations B.7 Preliminary wet lab validation License: arXiv.org perpetual non-exclusive license arXiv:2605.31236v1 [q-bio.BM] 29 May 2026 SwitchCraft : A Programmatic Framework for Designing State-Switching Proteins Bowen Jing Affiliation: CSAIL, MIT Correspondence to: [email protected] Mihir Bafna Affiliation: CSAIL, MIT Correspondence to: [email protected] Anisha Parsan Affiliation: CSAIL, MIT Heyuan Michael Ni Affiliation: Dept. of Biological Engineering, MIT David Kwabi-Addo Affiliation: CSAIL, MIT Bryan Bryson Affiliation: Dept. of Biological Engineering, MIT Adam Klivans Affiliation: Dept. of Computer Science, UT Austin Bonnie Berger Affiliation: CSAIL, MIT Affiliation: Dept. of Mathematics, MIT Correspondence to: [email protected] Abstract Multistate mechanisms underlie many of the complex functions observed in natural proteins. The ability to rationally design multistate proteins would have transformative implications for many areas of biotechnology, yet lies beyond the capabilities of existing deep learning frameworks for protein design. To address this gap, we introduce SwitchCraft , a versatile and programmatic framework for designing state-switching proteins based on backpropagation through compositional design constraints parameterized by structure prediction models. In silico evaluations demonstrate success on a wide range of state-switching functional primitives, from allosteric regulation of motifs to discrimination of bound ligand identities. Using these primitives, we demonstrate an in silico strategy for de novo design of fluorescent biosensors to arbitrary small molecule analytes. These results position SwitchCraft at the inception of a powerful paradigm for higher-order functional protein design. Co Hosted by Theo & Dr. Mara.

In recent years, scientists have developed methods to measure a cell's transcriptome, or all the RNA produced by a cell, to study the cell's identity and genetic activity. However, these methods rely on killing the cell to access the bits of RNA within and offer only a one-time snapshot. Hosted by Sofia & Daniel.

What's happening in Biology right now. Hosted by Sofia & Daniel.

Tracing the roots of Writing Genomes From Scratch across 6 papers. Hosted by Sofia & Daniel.

Bacterial promoters are DNA sequences that recruit RNA polymerase to initiate transcription, primarily through two elements called −10 and −35. However, the contributions of these elements to promoter evolution and regulation have remained unclear. Hosted by Theo & Dr. Mara.

For decades, scientists have understood that each of the roughly 20,000 genes in the human body carries instructions for a single kind of protein. Now, researchers at Harvard Medical School have discovered that instructions from different genes—even those on different chromosomes—can combine to create chimeric mRNAs that produce previously unknown, functional proteins. Hosted by Sofia & Daniel.

LPS-stimulated bone-marrow-derived macrophages make mitochondrial superoxide by reverse electron transport at complex I, driven by a raised membrane potential (because ATP production shifts to glycolysis and F1FO-ATP synthase runs less) and a CoQ pool kept reduced by succinate oxidation; a mouse carrying the ND6 G14600A (P25L) mtDNA variant, which blocks RET without affecting forward transport or complex III, makes no LPS-induced superoxide yet shows unchanged succinate, itaconate, lactate, NO, TNF, IL-6 and IL-10 — isolating RET-ROS as the specific regulator of IL-1beta release during NLRP3 activation. Hosted by Theo & Dr. Mara.

What's happening in AI right now. Hosted by Sofia & Daniel.

Tracing the roots of The Physics of Being Alive across 6 papers. Hosted by Theo & Dr. Mara.

In the ocean, plastic waste collects to form large garbage patches that endanger marine life. This is partly because the synthetic plastic polymers used today can only be broken down biologically—that is, by microorganisms—at a very slow rate. Instead, the physical fragmentation of these materials leads to the formation of microplastics and nanoplastics. However, microorganisms do colonize plastic waste in the environment, forming biofilms and creating a unique microbial habitat that researchers refer to as the "plastisphere." Hosted by Sofia & Daniel.