@pleaseignore1.bsky.social
Bottom-up design of calcium channels from defined selectivity filter geometry [new]
RFdiffusion bottom-up creates Ca2+ channels; filter-selectivity tested via construction.
RFdiffusion bottom-up creates Ca2+ channels; filter-selectivity tested via construction.
December 21, 2024 at 5:49 AM
Bottom-up design of calcium channels from defined selectivity filter geometry [new]
RFdiffusion bottom-up creates Ca2+ channels; filter-selectivity tested via construction.
RFdiffusion bottom-up creates Ca2+ channels; filter-selectivity tested via construction.
Cryo-EM was used to determine the structure of CalC6_3. To aid structural determination, a fusion protein strategy was employed by connecting a designed helical repeat to the C-terminus of CalC6_3. The structure of the hexamer was determined at 3.75 Å with C6 symmetry (Fig 5). The density for...
December 21, 2024 at 3:32 AM
Cryo-EM was used to determine the structure of CalC6_3. To aid structural determination, a fusion protein strategy was employed by connecting a designed helical repeat to the C-terminus of CalC6_3. The structure of the hexamer was determined at 3.75 Å with C6 symmetry (Fig 5). The density for...
Whole-cell patch-clamp experiments on Hi5 cells expressing the designed channels showed increased inward currents when extracellular Ca2+ was increased (Fig. 3b,c). The I-V curves exhibited inward rectification, consistent with the experimental setup where Ca2+ was predominantly extracellular...
December 21, 2024 at 3:32 AM
Whole-cell patch-clamp experiments on Hi5 cells expressing the designed channels showed increased inward currents when extracellular Ca2+ was increased (Fig. 3b,c). The I-V curves exhibited inward rectification, consistent with the experimental setup where Ca2+ was predominantly extracellular...
Functional channels were identified using a cell-based flux assay with Fura-2 AM and Ba2+ as a Ca2+ surrogate. Several designs in the CalC4, CalC6, and CalC6_H series showed increased Ba2+ flux. The designs that showed activity were expressed in E. coli, purified and analyzed by SEC and ns-EM....
December 21, 2024 at 3:32 AM
Functional channels were identified using a cell-based flux assay with Fura-2 AM and Ba2+ as a Ca2+ surrogate. Several designs in the CalC4, CalC6, and CalC6_H series showed increased Ba2+ flux. The designs that showed activity were expressed in E. coli, purified and analyzed by SEC and ns-EM....
The authors aimed to create Ca2+ channels by first defining the selectivity filter geometry, then building the protein scaffold around it. They systematically sampled the distances and coordination of carboxylate sidechains around a central Ca2+ ion (Fig. 1a). They used deep learning methods...
December 21, 2024 at 3:32 AM
The authors aimed to create Ca2+ channels by first defining the selectivity filter geometry, then building the protein scaffold around it. They systematically sampled the distances and coordination of carboxylate sidechains around a central Ca2+ ion (Fig. 1a). They used deep learning methods...
Guiding Generative Protein Language Models with Reinforcement Learning [new]
Fine-tuning protein language models via RL guides them towards rare, high-value sequences by maximizing custom reward signals; applied for EGFR binder design.
Fine-tuning protein language models via RL guides them towards rare, high-value sequences by maximizing custom reward signals; applied for EGFR binder design.
December 18, 2024 at 5:50 AM
Guiding Generative Protein Language Models with Reinforcement Learning [new]
Fine-tuning protein language models via RL guides them towards rare, high-value sequences by maximizing custom reward signals; applied for EGFR binder design.
Fine-tuning protein language models via RL guides them towards rare, high-value sequences by maximizing custom reward signals; applied for EGFR binder design.
The authors quantitatively assess pDIFF's ability to recapitulate phenotypic outcomes by calculating Spearman correlation coefficients between real and generated images (pg. 6). Specifically, they computed features such as cell coverage, cell count, and cell size. The improved correlation...
December 15, 2024 at 1:32 PM
The authors quantitatively assess pDIFF's ability to recapitulate phenotypic outcomes by calculating Spearman correlation coefficients between real and generated images (pg. 6). Specifically, they computed features such as cell coverage, cell count, and cell size. The improved correlation...
To test the utility of pDIFF in a virtual hit expansion, the authors compared the top-50 nearest neighbors using real images and pDIFF-generated images. The pDIFF-based retrieval method showed significant improvement in identifying compounds with similar phenotypes, with a 50% median...
December 15, 2024 at 1:32 PM
To test the utility of pDIFF in a virtual hit expansion, the authors compared the top-50 nearest neighbors using real images and pDIFF-generated images. The pDIFF-based retrieval method showed significant improvement in identifying compounds with similar phenotypes, with a 50% median...
compounds like CHEMBL2326002 and Brusatol, which the baseline model also struggled with (pg. 6). See Figure 2 for a comparison of real and generated images. Table 1 shows that pDIFF yields improved correlation coefficients across a range of image-derived features for the held-out compounds...
December 15, 2024 at 1:32 PM
compounds like CHEMBL2326002 and Brusatol, which the baseline model also struggled with (pg. 6). See Figure 2 for a comparison of real and generated images. Table 1 shows that pDIFF yields improved correlation coefficients across a range of image-derived features for the held-out compounds...
fingerprints, which limits generalizability to structurally similar compounds (pg. 2-3). By using bioactivity profiles as input to the diffusion model, the method is able to learn the "language of compound bioactivity" and extrapolate to novel compounds (pg. 4). See Figure 1 for a workflow...
December 15, 2024 at 1:31 PM
fingerprints, which limits generalizability to structurally similar compounds (pg. 2-3). By using bioactivity profiles as input to the diffusion model, the method is able to learn the "language of compound bioactivity" and extrapolate to novel compounds (pg. 4). See Figure 1 for a workflow...
December 13, 2024 at 4:20 PM
December 13, 2024 at 4:09 PM
December 13, 2024 at 4:07 PM
In hematopoiesis, RegVelo recovered five terminal states and recapitulated state transitions. In silico perturbations and GRN analysis identified key lineage regulators and a toggle switch motif between SPI1 and GATA1, consistent with previous studies (pg. 13, Figure 3).
December 12, 2024 at 5:35 PM
In hematopoiesis, RegVelo recovered five terminal states and recapitulated state transitions. In silico perturbations and GRN analysis identified key lineage regulators and a toggle switch motif between SPI1 and GATA1, consistent with previous studies (pg. 13, Figure 3).
Applied to pancreatic endocrine development, RegVelo delineated ductal cell cycling and predicted four terminal states. In silico perturbations identified a ductal cell subpopulation and confirmed E2f1's role in cell cycling regulation (pgs. 9-10, Figure 2).
December 12, 2024 at 5:35 PM
Applied to pancreatic endocrine development, RegVelo delineated ductal cell cycling and predicted four terminal states. In silico perturbations identified a ductal cell subpopulation and confirmed E2f1's role in cell cycling regulation (pgs. 9-10, Figure 2).
RegVelo uses a Bayesian deep generative model to describe cell dynamics. It models unspliced/spliced RNA readouts using kinetic & neural network parameters, capturing how upstream regulators control gene transcription (pg. 4).
December 12, 2024 at 5:35 PM
RegVelo uses a Bayesian deep generative model to describe cell dynamics. It models unspliced/spliced RNA readouts using kinetic & neural network parameters, capturing how upstream regulators control gene transcription (pg. 4).
In hematopoiesis, RegVelo recovered five terminal states and recapitulated state transitions. In silico perturbations and GRN analysis identified key lineage regulators and a toggle switch motif between SPI1 and GATA1, consistent with previous studies (pg. 13, Figure 3).
December 12, 2024 at 5:33 PM
In hematopoiesis, RegVelo recovered five terminal states and recapitulated state transitions. In silico perturbations and GRN analysis identified key lineage regulators and a toggle switch motif between SPI1 and GATA1, consistent with previous studies (pg. 13, Figure 3).
Applied to pancreatic endocrine development, RegVelo delineated ductal cell cycling and predicted four terminal states. In silico perturbations identified a ductal cell subpopulation and confirmed E2f1's role in cell cycling regulation (pgs. 9-10, Figure 2).
December 12, 2024 at 5:33 PM
Applied to pancreatic endocrine development, RegVelo delineated ductal cell cycling and predicted four terminal states. In silico perturbations identified a ductal cell subpopulation and confirmed E2f1's role in cell cycling regulation (pgs. 9-10, Figure 2).
RegVelo uses a Bayesian deep generative model to describe cell dynamics. It models unspliced/spliced RNA readouts using kinetic & neural network parameters, capturing how upstream regulators control gene transcription (pg. 4).
December 12, 2024 at 5:33 PM
RegVelo uses a Bayesian deep generative model to describe cell dynamics. It models unspliced/spliced RNA readouts using kinetic & neural network parameters, capturing how upstream regulators control gene transcription (pg. 4).
Ablation studies show a positive correlation between the number of replaced MLP layers with LKANs and the model's performance (pg. 6). Increasing the grid size parameter also improves performance but increases the model size and training time (Figure 3). (pg. 5-6, Figure 3)
December 11, 2024 at 4:16 PM
Ablation studies show a positive correlation between the number of replaced MLP layers with LKANs and the model's performance (pg. 6). Increasing the grid size parameter also improves performance but increases the model size and training time (Figure 3). (pg. 5-6, Figure 3)
For sequence generation, LKANs are incorporated into a DNA-Diffusion U-Net model. Figure 2 shows the architecture of the U-Net with LKAN modifications. Results demonstrate LKAN's ability to achieve lower validation loss compared to the baseline U-Net and CKANs with similar or fewer parameters...
December 11, 2024 at 4:16 PM
For sequence generation, LKANs are incorporated into a DNA-Diffusion U-Net model. Figure 2 shows the architecture of the U-Net with LKAN modifications. Results demonstrate LKAN's ability to achieve lower validation loss compared to the baseline U-Net and CKANs with similar or fewer parameters...
LKANs consistently outperform both the baseline and CKAN models across nearly all datasets in the classification tasks (Tables 1-3). CKANs struggle with scaling to larger parameter counts. (pg. 5, Table 1, Table 2, Table 3)
December 11, 2024 at 4:16 PM
LKANs consistently outperform both the baseline and CKAN models across nearly all datasets in the classification tasks (Tables 1-3). CKANs struggle with scaling to larger parameter counts. (pg. 5, Table 1, Table 2, Table 3)
For sequence classification, both LKAN and CKAN architectures are integrated into a LegNet-based CNN. Figure 1 illustrates the architectures for the baseline LegNet, the LKAN-modified LegNet, and the CKAN-modified LegNet. (pg. 4, Figure 1)
December 11, 2024 at 4:16 PM
For sequence classification, both LKAN and CKAN architectures are integrated into a LegNet-based CNN. Figure 1 illustrates the architectures for the baseline LegNet, the LKAN-modified LegNet, and the CKAN-modified LegNet. (pg. 4, Figure 1)